n° d ordre 4663 - u-bordeaux1.frori-oai.u-bordeaux1.fr/pdf/2012/medjiah_samir_2012.pdf · plus...
TRANSCRIPT
i
N° D'ORDRE : 4663
THÈSE
présentée à
L'UNIVERSITÉ BORDEAUX 1
École Doctorale de Mathématiques et d'Informatique
Par
Samir Medjiah
Pour obtenir le grade de
DOCTEUR
SPÉCIALITÉ: INFORMATIQUE
ROUTING PROTOCOL OPTIMIZATION IN CHALLENGED MULTIHOP
WIRELESS NETWORKS
OPTIMISATION DES PROTOCOLES DE ROUTAGE DANS LES RÉSEAUX
MULTI-SAUTS SANS FIL À CONTRAINTES.
Soutenue le: 10 Décembre 2012
Devant la commission d'examen composée de:
M. AHMED Toufik Professeur, ENSEIRB-MATMECA – IPB Directeur de thèse
M. GHAMRI-DOUDANE Yacine Maitre de Conférences (HDR), ENSEIIE Rapporteur
M. LAGRANGE Xavier Professeur, Télécom Bretagne Président du jury
M. MEDDOUR Djamal-Eddine Responsable R&D, Orange Labs Examinateur
M. MOSBAH Mohamed Professeur, ENSEIRB-MATMECA – IPB Examinateur
M. PHAM Congduc Professeur, Université de Pau Rapporteur
ii
iii
Abstract
Great research efforts have been carried out in the field of challenged multihop wireless
networks (MWNs). Thanks to the evolution of the Micro-Electro-Mechanical Systems (MEMS)
technology and nanotechnologies, multihop wireless networks have been the solution of choice for
a plethora of problems. The main advantage of these networks is their low manufacturing cost that
permits one-time application lifecycle. However, if nodes are low-costly to produce, they are also
less capable in terms of radio range, bandwidth, processing power, memory, energy, etc.
Thus, applications need to be carefully designed and especially the routing task because radio
communication is the most energy-consuming functionality and energy is the main issue for
challenged multihop wireless networks.
The aim of this thesis is to analyse the different challenges that govern the design of
challenged multihop wireless networks such as applications challenges in terms of quality of service
(QoS), fault-tolerance, data delivery model, etc., but also networking challenges in terms of
dynamic network topology, topology voids, etc. Our contributions in this thesis focus on the
optimization of routing under different application requirements and network constraints. First,
we propose an online multipath routing protocol for QoS-based applications using wireless
multimedia sensor networks. The proposed protocol relies on the construction of multiple paths
while transmitting data packets to their destination, i.e. without prior topology discovery and path
establishment. This protocol achieves parallel transmissions and enhances the end-to-end
transmission by maximizing path bandwidth and minimizing the delays, and thus meets the
requirements of QoS-based applications. Second, we tackle the problem of routing in mobile delay-
tolerant networks by studying the intermittent connectivity of nodes and deriving a contact model
in order to forecast future nodes' contacts. Based upon this contact model, we propose a routing
protocol that makes use of nodes' locations, nodes' trajectories, and inter-node contact prediction
in order to perform forwarding decisions. The proposed routing protocol achieves low end-to-end
delays while using efficiently constrained nodes' resources in terms of memory (packet queue
occupancy) and processing power (forecasting algorithm). Finally, we present a topology control
mechanism along a packet forwarding algorithm for event-driven applications using stationary
wireless sensor networks. Topology control is achieved by using a distributed duty-cycle scheduling
algorithm. Algorithm parameters can be tuned according to the desired node's awake
neighbourhood size. The proposed topology control mechanism ensures trade-off between event-
reporting delay and energy consumption.
Key-words: WSN, WMSN, DTN, MDTN, Routing, Multipath, Energy-Aware, Contact
Forecasting, Geocast, Trajectory-Assistance, Topology Control, Duty-cycle scheduling.
iv
Résumé
Durant ces dernières années, de nombreux travaux de recherches ont été menés dans le
domaine des réseaux multi-sauts sans fil à contraintes (MWNs: Multihop Wireless Networks).
Grâce à l'évolution de la technologie des systèmes mico-electro-méchaniques (MEMS) et, depuis
peu, les nanotechnologies, les MWNs sont une solution de choix pour une variété de problèmes. Le
principal avantage de ces réseaux est leur faible coût de production qui permet de développer des
applications ayant un unique cycle de vie. Cependant, si le coût de fabrication des nœuds
constituant ce type de réseaux est assez faible, ces nœuds sont aussi limités en capacité en termes de:
rayon de transmission radio, bande passante, puissance de calcul, mémoire, énergie, etc. Ainsi, les
applications qui visent l'utilisation des MWNs doivent être conçues avec une grande précaution, et
plus spécialement la conception de la fonction de routage, vu que les communications radio
constituent la tâche la plus consommatrice d'énergie.
Le but de cette thèse est d'analyser les différents défis et contraintes qui régissent la conception
d'applications utilisant les MWNs. Ces contraintes se répartissent tout le long de la pile
protocolaire. On trouve au niveau application des contraintes comme: la qualité de service, la
tolérance aux pannes, le modèle de livraison de données au niveau application, etc. Au niveau
réseau, on peut citer les problèmes de la dynamicité de la topologie réseau, la présence de trous, la
mobilité, etc. Nos contributions dans cette thèse sont centrées sur l'optimisation de la fonction de
routage en considérant les besoins de l'application et les contraintes du réseau.
Premièrement, nous avons proposé un protocole de routage multi-chemin "en ligne" pour les
applications orientées QoS utilisant des réseaux de capteurs multimédia. Ce protocole repose sur la
construction de multiples chemins durant la transmission des paquets vers leur destination, c'est-à-
dire sans découverte et construction des routes préalables. En permettant des transmissions
parallèles, ce protocole améliore la transmission de bout-en-bout en maximisant la bande passante
du chemin agrégé et en minimisant les délais. Ainsi, il permet de répondre aux exigences des
applications orientées QoS.
Deuxièmement, nous avons traité le problème du routage dans les réseaux mobiles tolérants
aux délais. Nous avons commencé par étudier la connectivité intermittente entre les différents et
nous avons extrait un modèle pour les contacts dans le but pouvoir prédire les future contacts entre
les nœuds. En se basant sur ce modèle, nous avons proposé un protocole de routage, qui met à
profit la position géographique des nœuds, leurs trajectoires, et la prédiction des futurs contacts
dans le but d'améliorer les décisions de routage. Le protocole proposé permet la réduction des délais
v
de bout-en-bout tout en utilisant d'une manière efficace les ressources limitées des nœuds que ce soit
en termes de mémoire (pour le stockage des messages dans les files d'attentes) ou la puissance de
calcul (pour l'exécution de l'algorithme de prédiction).
Finalement, nous avons proposé un mécanisme de contrôle de la topologie avec un algorithme
de routage des paquets pour les applications orientés évènement et qui utilisent des réseaux de
capteurs sans fil statiques. Le contrôle de la topologie est réalisé à travers l'utilisation d'un
algorithme distribué pour l'ordonnancement du cycle de service (sleep/awake). Les paramètres de
l'algorithme proposé peuvent être réglés et ajustés en fonction de la taille du voisinage actif désiré
(le nombre moyen de voisin actifs pour chaque nœud). Le mécanisme proposé assure un compromis
entre le délai pour la notification d'un évènement et la consommation d'énergie globale dans le
réseau.
Mots-clés: réseaux de capteurs sans fil, réseaux de capteurs multimédia, réseaux tolérants
aux délais, réseaux mobiles, routage, routage multi-chemin, conservation d'énergie, prédiction
des contacts, routage géographique, contrôle de la topologie, ordonnancement de l'activité du
nœud.
vi
Table of Contents
Abstract ................................................................................................................................. iii
Résumé .................................................................................................................................. iv
Table of Contents .................................................................................................................. vi
List of Figures ........................................................................................................................ ix
List of Tables ........................................................................................................................ xii
Acknowledgements ............................................................................................................... xiv
1 Introduction .................................................................................................................... 1
1.1 Context and Motivation ............................................................................................. 1
1.2 Research Problem ....................................................................................................... 2
1.3 Contributions ............................................................................................................. 4
1.4 Thesis Structure .......................................................................................................... 6
2 Challenged Multihop Wireless Networks ....................................................................... 8
2.1 Potential Applications Overview ................................................................................ 9
2.2 Wireless Node Architecture ...................................................................................... 10
2.3 Low Cost Multihop Wireless Networks Characteristics ........................................... 11
2.4 Challenges and Design Principles .............................................................................. 12
2.4.1 Fault-Tolerance ................................................................................................. 12
2.4.2 Scalability .......................................................................................................... 12
2.4.3 Manufacturing Cost ........................................................................................... 13
2.4.4 Hardware Constraints ....................................................................................... 13
2.4.5 Physical Environment ....................................................................................... 13
2.4.6 Network Topology ........................................................................................... 13
2.4.7 Energy Consumption ........................................................................................ 13
2.5 Challenges across the Communication Protocols Stack ............................................ 14
2.5.1 Applications Level Challenges ........................................................................... 14
2.5.2 Network Level Challenges ................................................................................ 15
2.5.3 Physical Level Challenges .................................................................................. 16
2.6 Analysis of Routing protocols for MWNs ................................................................ 17
2.6.1 Network structure based routing protocols ...................................................... 17
2.6.2 Protocol operation-based routing protocols ...................................................... 18
2.7 Delay-Tolerant Routing in MWNs ........................................................................... 19
2.7.1 DTN Characteristics ......................................................................................... 20
2.7.2 Routing Problem in DTNs ............................................................................... 21
2.7.3 Routing Topologies for DTNs .......................................................................... 22
2.7.4 Data Packet Forwarding Strategies .................................................................... 23
2.8 Self-Organization in MWNs ..................................................................................... 27
vii
2.8.1 Link pruning techniques ................................................................................... 27
2.8.2 Dominating Set Techniques .............................................................................. 30
2.9 Conclusion ................................................................................................................ 32
3 Online Multipath Routing in Multi-hop Wireless Networks ....................................... 34
3.1 Introduction .............................................................................................................. 34
3.2 Related Work ............................................................................................................ 37
3.2.1 The GPSR Routing Protocol ............................................................................. 37
3.2.2 The TPGF Routing Protocol ............................................................................ 38
3.2.3 The MPMPS Routing Protocol ......................................................................... 38
3.2.4 Policies for Greedy forwarding ......................................................................... 39
3.2.5 Discussion on Routing/Forwarding .................................................................. 40
3.3 AGEM Routing Protocol.......................................................................................... 40
3.3.1 Smart Greedy forwarding mode: ....................................................................... 41
3.3.2 Walking Back forwarding mode ........................................................................ 45
3.4 Performances Evaluation .......................................................................................... 47
3.4.1 Simulation Environment ................................................................................... 47
3.4.2 Simulation Results: ............................................................................................ 50
3.4.3 Simulation Results Discussion ........................................................................... 54
3.5 Conclusion ................................................................................................................ 56
4 Predictive Routing in Mobile Wireless Networks ........................................................ 57
4.1 Introduction .............................................................................................................. 57
4.2 Related Work ............................................................................................................ 58
4.2.1 DTN Routing Protocols Taxonomy ................................................................. 59
4.2.2 Times Series in Network Modelling .................................................................. 60
4.3 ORION Routing Protocol ........................................................................................ 62
4.3.1 Target Application ............................................................................................ 62
4.3.2 Contact Behaviour Analysis .............................................................................. 63
4.3.3 ORION Contact Model Construction .............................................................. 64
4.3.4 Forwarding Algorithm ...................................................................................... 69
4.4 Performances Evaluation .......................................................................................... 71
4.4.1 Simulation Environment ................................................................................... 71
4.4.2 Simulation Results Discussion ........................................................................... 71
4.5 Conclusion ................................................................................................................ 74
5 Stochastic Topology Control in Wireless Sensor Networks ......................................... 75
5.1 Introduction .............................................................................................................. 75
5.2 Related Work ............................................................................................................ 76
5.2.1 Sensor Coverage Topology................................................................................ 76
5.2.2 Sensor Connectivity Topology ......................................................................... 77
viii
5.2.3 Discussion ......................................................................................................... 77
5.3 System Model ............................................................................................................ 78
5.3.1 Node Deployment............................................................................................. 79
5.3.2 Wakeup/Sleep Schedule .................................................................................... 80
5.3.3 Target Wakeup/Sleep Schedule ......................................................................... 82
5.3.4 Local Topology Awareness and Path Construction .......................................... 85
5.3.5 Data Packet Forwarding .................................................................................... 87
5.4 Performances Evaluation .......................................................................................... 88
5.4.1 Simulation Environment ................................................................................... 88
5.4.2 Simulation Results ............................................................................................. 89
5.5 Conclusion ................................................................................................................ 94
6 Conclusion and Perspectives ......................................................................................... 96
6.1 Thesis Contributions ................................................................................................ 96
6.2 Perspectives ............................................................................................................... 98
Publications ........................................................................................................................... 99
Journals .............................................................................................................................. 99
International Conferences .................................................................................................. 99
National Conferences......................................................................................................... 99
Workshops ......................................................................................................................... 99
References ............................................................................................................................ 100
ix
List of Figures
Figure 1: Multihop Wireless Network Architecture ................................................................ 8
Figure 2: Wireless Node Architecture .................................................................................... 10
Figure 3: Presence of Void in the Sensing Field ..................................................................... 16
Figure 4: Taxonomy of Routing Protocols for WSNs. .......................................................... 17
Figure 5: DTN Routing Strategies Classification ................................................................... 23
Figure 6: Link Pruning Mechanism in RNG ......................................................................... 28
Figure 7: Link Pruning Mechanism in GG ............................................................................ 28
Figure 8: Delaunay Graph along Voronoï Diagram ............................................................... 29
Figure 9: Link Pruning Mechanism in LMST. ....................................................................... 29
Figure 10: Initial Unit Disc Graph[103] ................................................................................. 31
Figure 11: Gabriel Graph (GG) [103] ..................................................................................... 31
Figure 12: Relative Neighbour Graph (RNG) [103] ............................................................... 31
Figure 13: Local Minimum Spanning Tree (LMST) [103] ...................................................... 31
Figure 14: Initial Unit Disc Graph[102] ................................................................................. 32
Figure 15: Connected Dominating Set[102] ........................................................................... 32
Figure 16: GPSR Perimeter Forwarding to Bypass a Void. .................................................... 38
Figure 17: Greedy Forwarding Strategies: (a) Compass Routing; (b) Random Compass
Routing; (c) Greedy Routing; (d) Most Forwarding; (e) Nearest Neighbour Routing; (f) Furthest
Neighbour Routing. ......................................................................................................................... 40
Figure 18: GEAMS Routing Mode Switching. ....................................................................... 41
Figure 19: AGEM Adaptive Compass Policy. ........................................................................ 42
Figure 20: One-hop Neighbours Sorted According to their Scores. ....................................... 43
Figure 21: Packet Energy Consumption between two Communicating Nodes A and B. ...... 43
Figure 22: The Smart Greedy Forwarding Algorithm. .......................................................... 45
Figure 23: Forwarding the First Packet of a Data Stream. ..................................................... 46
Figure 24: Forwarding a Packet of an already Known Data Stream. ...................................... 46
Figure 25: A Blocking Situation where a Node has no Forwarder Node. .............................. 47
Figure 26: Data Delivery in Response to an Event in a WMSN. ............................................ 47
Figure 27: A 30-nodes network topology. .............................................................................. 49
Figure 28: A 30-nodes network topology with two holes. ..................................................... 49
Figure 29: A 26-nodes grid network topology. ...................................................................... 49
Figure 30: Average Residual Energy in “Plain” Topologies. .................................................. 50
Figure 31: Residual Energy Distribution for 30-Node Network Topology ........................... 51
x
Figure 32: Residual Energy Distribution or 50-Node Network Topology. ........................... 51
Figure 33: Residual Energy Distribution for 80-Nodes Topology. ........................................ 51
Figure 34: Average end-to-end delay in plain topologies. ....................................................... 52
Figure 35: Packet-loss ratio in plain topologies. ..................................................................... 52
Figure 36: Average residual energy in topologies with holes. ................................................. 53
Figure 37: Residual Energy Distribution across the Network for 50-Node Network
Topology with two holes. ................................................................................................................ 53
Figure 38: Average End-to-End Delay in Topologies with Holes. ......................................... 53
Figure 39: The Packet-loss Ratio in Topologies with Holes (Logarithmic Scale) ................... 54
Figure 40: Residual Energy with GPSR in a Grid Topology. ................................................ 54
Figure 41: Residual Energy with AGEM in a Grid Topology. .............................................. 54
Figure 42: Variation of and over time. ......................................................................... 63
Figure 43: Autocorrelation Function (ACF) plot for contact's duration. .............................. 64
Figure 44: Partial Autocorrelation Function (PACF) plot for contact's duration. ................. 64
Figure 45: Online model parameters update, and future value forecasting ............................. 68
Figure 46: Forecasting with "online" and "offline" parameter estimation. ............................. 69
Figure 47: Pseudo code for ORION forwarding algorithm. .................................................. 69
Figure 48: A detailed description of the ORION forwarding algorithm. .............................. 70
Figure 49: Average hop count in 30 and 70 nodes topologies with variant speed. .................. 71
Figure 50: Packet Success Ratio in 30 and 70 nodes topologies with variant speed. ............... 72
Figure 51: First Packet Arrival in 30 and 70 nodes topologies with variant speed. ................ 72
Figure 52: Average E2E Delay in 30 and 70 nodes topologies with variant speed. ................. 73
Figure 53: Maximum message queue occupancy in 30 nodes topologies with variant speed. . 73
Figure 54: Transmission over Long Distances ........................................................................ 75
Figure 55: Hot Spot Node...................................................................................................... 75
Figure 56: Neighbouring Probability in Uniform Deployment ............................................. 79
Figure 57: Number of Active Neighbours ............................................................................. 80
Figure 58: Probability to have K neighbours ......................................................................... 83
Figure 59: Probability to Have [u,v] Active Neighbours ....................................................... 84
Figure 60: Path Construction Algorithm ............................................................................... 86
Figure 61: Example of Path Construction .............................................................................. 86
Figure 62: Example of Path Recovery .................................................................................... 87
Figure 63: Packet Forwarding Algorithm .............................................................................. 87
Figure 64: Average Awake Neighbours in 100 Nodes Network ............................................ 89
Figure 65: Average Awake Neighbours in 150 Nodes Network ............................................ 90
xi
Figure 66: Average Awake Neighbours in 200 Nodes Network ............................................ 90
Figure 67: Average Awake Neighbours in 250 Nodes Network ............................................ 90
Figure 68: Average Sleep Duration (%) of the Network ........................................................ 91
Figure 69: Average End-to-End Delay (sec) with 100s Average Event Generation Interval ... 92
Figure 70: Average End-to-End Delay (sec) with 150s Average Event Generation Interval ... 92
Figure 71: Average End-to-End Delay (sec) with 200s Average Event Generation Interval ... 92
Figure 72: Average Wait-to-Send Delay (sec) with 100s Average Event Generation Interval . 93
Figure 73: Average Wait-to-Send Delay (sec) with 150s Average Event Generation Interval . 93
Figure 74: Average Wait-to-Send Delay (sec) with 200s Average Event Generation Interval . 93
Figure 75: Average Queue Occupancy (unit) with 100s Average Event Generation Interval 94
Figure 76: Average Queue Occupancy (unit) with 150s Average Event Generation Interval 94
Figure 77: Average Queue Occupancy (unit) with 200s Average Event Generation Interval 94
xii
List of Tables
Table 1: Simulation Parameters (WMSNs). ............................................................................ 48
Table 2: Simulation Parameters (Topology Control) ............................................................. 88
Table 3: Simulated Values for Wakeup/Sleep Ratio ............................................................... 89
xiii
In memory of my dear father,
To my dear mother, brother and sister,
To my uncle Boualem who was always there for every important decision of my life.
To my fiancée Fatima and all my family,
I dedicate this work.
xiv
Acknowledgements
First and foremost, I would like to greatly thank my advisor Prof. Toufik AHMED who
guided me throughout my thesis work. Prof. Toufik Ahmed's dedication, advice, perseverance and
guidance have made this thesis possible. He has been an excellent motivator.
I am grateful to Prof. Congduc PHAM and Dr. Yacine GHAMRI-DOUDANE for accepting
to review and evaluate this thesis. Their pertinent remarks about my research work have been of
great profit for me. I am also very thankful to Prof. Xavier LAGRANGE, Prof. Mohamed
MOSBAH, and Dr. Djamal-Eddine MEDDOUR for accepting to be the examiners of my thesis
defence.
I would like to thank my colleagues within the COMET-MUSE team at LaBRI Laboratory
for kindly sharing the experience on their study and work. I would also like to thank all my friends
for their unlimited support and encouragement during all the preparation of thesis.
Finally, I am thankful to all those who directly or indirectly contributed to the realization
of this work, for their help, encouragement or simply by their presence.
1
Chapter 1
1 Introduction
1.1 Context and Motivation
Great research efforts have been carried out in the field of low-cost multihop wireless
networks (MWNs), especially wireless sensor networks (WSNs). Thanks to the evolution of the
Micro-Electro-Mechanical Systems (MEMS) technology and nanotechnologies, these challenged
networks have been the solution of choice for a plethora of problems.
The main advantage of these networks is their relative low manufacturing cost. Thanks to
this advantage, applications developers could offer the comfort of one-time application lifecycle.
Indeed, such networks' nodes can be dropped and left working on batteries without worrying to
replace their batteries or even retrieving them from the deployment field. However, the other side
of the coin is that if the network nodes are low-costly to produce, they are also less capable in terms
of radio transmission range, bandwidth capacity, processing power, memory storage, etc. Due to
these limitations, applications that use low cost MWNs need to be carefully designed and classic
approaches that are valid for today's networks (i.e. AdHoc Networks) do not apply for these
networks. Every transmitted bit of data, every executed processor instruction, and every stored
byte need to be totally justified and necessary.
Many research works can be found in the literature where researchers have used challenged
multihop wireless networks for various applications (natural phenomenon motoring, data gathering
and dissemination, etc.). Most of these research works did not consider the minimal capabilities of
these networks as a stringent constraint. Such inefficiency can be witnessed through routing
protocols based on massive data packet exchange for topology discovery or path construction.
These protocols have been inherited from the advances achieved in the field of ad-hoc wireless
networks. Another inefficiency can also be noticed when it comes to some algorithms run on the
wireless nodes. A large number of research works developed CPU-demanding data processing
algorithms, thus, the carried energy will deplete quickly. Therefore, this comes in contradiction
with the philosophy behind low cost multihop wireless networks.
This thesis was undertaken with the strong belief that challenged MWNs are not costly to
manufacture but they are very limited. Thus, solutions that can be proposed must be as optimized
as possible to be in line with the idea that low cost MWNs can be used to achieve an acceptable
2
service as good as what legacy infrastructures can achieve and without making the wireless nodes
more expensive to produce by putting more energy, enhanced radio communications hardware,
powerful CPUs, or any additional hardware (i.e., mobility components).
1.2 Research Problem
The works that have been carried out in this thesis are oriented to the multihop wireless
networks (MWNs) in general but especially to the most challenged networks among them
including wireless sensor networks (WNSs), wireless multimedia sensor networks (WMSNs), and
mobile delay-tolerant networks (MDTNs). In order to present a clear scope of the work, we
introduce the challenges that are related to our studied subject.
Low-cost multihop wireless networks are composed of a large number of nodes which are
small in size. These nodes are capable of sensing data from the deployment environments,
processing the data and communicating the results to the end user or the control entity. They can
also communicate with each other and cooperate in order to achieve a determined task. In addition
to the inherited characteristics related to wireless communications, low cost multihop wireless
networks have other characteristics inherent to the way they operate:
— Limited capabilities. Nodes often operate on (irreplaceable) batteries with: limited energy,
transmit power, memory and computing capabilities.
— Short range radio communications. Network nodes use multihop communications in
order to transmit data for long distances.
— Dynamic network topology. Network topology may be subject to frequent changes due to
nodes' failure, nodes' mobility or any energy conservation mechanism run by the node
to adjust the transmitting power or to schedule the node activity (wakeup/sleep
scheduling).
— Data centric communications model. Collected data are more important than the source
nodes identities (unless the node identifier actually carries some information such as the
location where the data has been measured which is the case when geographical node
addressing is used).
— Dense and redundant node deployment. In order to leverage nodes' failures and to increase
the overall network lifetime, nodes are often deployed densely and redundantly.
— Need for self-* functionalities (organization, control, healing, etc.). Since wireless nodes are
often left unattended, such functionalities need to be taken into consideration during the
design steps of the applications.
3
Low cost MWNs have been used for various applications in different fields: military and
defence (border monitoring, intrusion detection, video surveillance, etc.), civil (infrastructure health
monitoring), natural environments (forest fire, volcano activity, oceanography, etc.), commerce (in-
door localization, precision farming, traffic control, etc.), medical (disaster prevention, medical
care, etc.). It is easy to notice that these applications have different needs in terms of transmitted
data volumes, data rates, reporting frequencies, etc. Some of these applications require the
transmission of large quantities of data (especially when it comes to multimedia contents such as
video, audio, or even a basic image stream). Sending such data streams through low cost MWNs
may be very difficult, even impossible in certain situations. The main difficulty is due to the low
bandwidth available in each hop. To overcome this problem, the use of multiple path transmission
seems of good choice. However, the construction of multiple paths for parallel transmission
identifies some challenges such as network topology awareness, path building and maintenance, and
finally load-balancing. Since such applications put a heavy load on actual data transmission,
overburdening these data communications with control packets is inefficient as done by the huge
majority of research works in the field of multipath routing protocols for wireless sensor networks.
Thus, an interesting challenge in such configuration is an online multipath routing protocol; a
routing protocol where the establishment, maintenance and load-balancing of multiple parallel
paths is achieved without a complete knowledge of the network topology.
In other applications where nodes are densely deployed across the sensing field but the
collected data is not frequent and not bandwidth demanding (such as intrusion detection or
environment monitoring applications), the data communications are event-driven. In this case, it is
interesting to not use the entire network nodes to achieve such service. It is customary to put the
maximum number of wireless nodes in a low consuming energy state (a sleep state) without
degrading the desired service in order to achieve energy efficiency and thus maximizing the
network lifetime. Such solutions will result in a dynamic network topology due to nodes'
wakeup/sleep states switching. Therefore, the challenge is to design wakeup/sleep scheduling
algorithm that maintains a routing infrastructure capable of routing data packets from any source
towards the destination. Along this scheduling algorithm, and to reiterate our vision on low cost
MWNs, network topology awareness, construction and maintenance do not need to be energy,
memory or power consuming.
In a more advanced scenario, certain application can make use of mobile wireless nodes. If
the interest area becomes larger, nodes may be disconnected from each other for a long period of
time resulting in intermittent connectivity between network nodes and thus, a highly dynamic
network topology. Such infrastructure is useless for a majority of applications unless those that
4
tolerate relatively large end-to-end delays due to the store-carry-and-forward mechanism used to
forward data packets from the source to the destination. In this context, the efficiency of the
routing protocol depends heavily on the amount of knowledge the mobile node can acquire before
performing its routing decisions. The challenge behind the design of an efficient routing protocol
for mobile multihop wireless networks is first, the guarantee for a source to reach the destination,
and second, the minimization of the end-to-end delay. There are other secondary challenges such as
the resources utilization. Indeed, some routing protocols propose to duplicate the data packets
(replication-based protocols) in order to increase the probability of meeting the destination node
earlier. These protocols excessively fill intermediate nodes' queues with duplicated data packets
which will lead to inefficient resource utilization. In order to overcome this challenge, any other
knowledge about the network topology becomes beneficial and can enhance the routing protocol.
Examples of additional knowledge include:
— The use of geographical addresses permits the nodes to determine the trajectory of a node
during a contact. Thus, this information can help to make a decision about whether to
forward the packet to this node or not.
— The analysis of the inter-nodes encounters history can ease the forecast of the next
contact. More information can be forecasted too, such as next contact date, next contact
duration, etc.
Other challenges can be found in every specific application. However, the enumeration of all
the possible challenges is out of the scope of this thesis.
1.3 Contributions
During this thesis, we focused on some routing challenges for Multihop Wireless Networks.
In this scope, the main goal of this thesis is to study different routing issues and to propose adequate
solutions. Towards this goal, the problem of QoS-based applications over wireless sensor networks
has been treated by the proposition of an online multipath routing protocol. The problem of
intermittent connectivity and end-to-end delay minimization in the case of mobile delay-tolerant
networks has been tackled by the proposition of a model of inter-node contact forecasting and
forwarding-based routing protocol (i.e. only a single copy of the data packet exists in the entire
network). Finally, we have addressed the problem of the changing topology resulting from
wakeup/sleep scheduling (mechanism that aims to minimize the energy consumption) by the
proposition of a scheduling algorithm along topology control mechanisms.
The main contributions of this thesis can be summarized as follows:
5
— Online multipath routing protocol for WMSNs: our first contribution targeted QoS-based
applications such as real-time applications using densely deployed wireless multimedia
sensor networks, where nodes are equipped with audio / visual sensors. Such applications
generate high volumes of data that need to be transmitted across low capacity links in a
hop-by-hop basis. In order to not drain the nodes energy quickly, our proposed routing
protocol relies on geographical node addressing and greedy packet forwarding in order to
establish multiple paths without the knowledge of the entire network topology (only
one-hop neighbourhood awareness is used). Unlike offline multipath protocols where
paths building phase precedes data transmission phase, our routing protocol establishes
multiple paths during the data transmission, i.e. as data packets advance from the source
to the destination, they are forwarded across different paths. These multiple paths allow
parallel transmissions, and thus, achieve lower end-to-end delays and load-balancing since
the heavy load of high volumes of data transmissions becomes spread over a large
number of nodes contrary to single-path routing protocols.
— Inter-node contact model & Forwarding-based routing protocol for MDTNs: Our second
contribution focused on routing in delay tolerant networks and especially mobile DTNs.
In such networks, nodes exhibit intermittent connectivity resulting in a highly dynamic
network topology. We have considered a city-wide mobile network composed of
different nodes regarding their mobility model: (1) regular nodes represented by the
mobility of buses or trams through the use of defined time schedules and routes, (2)
irregular nodes represented by the mobility of users' cars, taxis, etc. having random
routes and schedules, and (3) static nodes represented by hotspots across the city.
Initially, we have studied the inter-node behaviour in this context and derived an inter-
node contact model that is based upon time series analysis. In a secondary step, we
proposed a routing protocol that makes use of the derived contact model for forecasting
next inter-node encounters, but it also makes use of knowledge about nodes locations and
trajectories in order to perform routing decisions. We have showed that our routing
protocol does not require either intensive computation for contact forecasting or
important memory space for packets buffering in order to achieve the minimization of
the end-to-end delays.
— Wakeup/Sleep scheduling algorithm & Topology control mechanisms for WSNs: our third
contribution focused on energy efficient routing infrastructure for dense WSNs used for
event-driven applications. We have proposed a distributed wakeup/sleep scheduling
algorithm that can be tuned in order to prune the active neighbourhood to a desired
average neighbourhood size. This restricted active neighbourhood permits the
6
economization of nodes' energy by letting various nodes to enter a low energy-
consuming state (sleep state) without degrading the data forwarding process. We have
also proposed additional mechanisms for routing path establishment and maintenance in
order to route data packets efficiently.
1.4 Thesis Structure
This thesis is organised as follows:
Chapter 2: reviews the challenges and considerations that govern routing protocols for
multihop wireless networks with a certain focus on the end-applications requirements and the
different issues that these applications face at different levels of the communications stack. It gives a
brief overview of the targeted applications and presents a classification of the different issues and
challenges tackled by existing routing protocols. An overview of routing protocols for WSNs is
presented. This chapter also discusses the problem of routing in DTNs by introducing these
networks, their characteristics, and a brief overview and classification of existing routing protocols
for DTNs. Finally, techniques for multihop wireless networks self-organization are also discussed.
Chapter 3: presents the details of our first contribution: an online multipath routing protocol
for wireless multimedia sensor networks for a QoS-based application. We first review some routing
protocols that influenced the design of our routing protocol such as geographical routing protocols
but also offline multipath protocols for WMSNs. Then, we describe our routing protocol
components: neighbour selection, packet forwarding scheme, load balancing, and topology holes
avoidance. The proposed solution takes into account both residual energy and packets transmission
delay in order to perform routing decisions.
Chapter 4: presents the details of our second contribution namely ORION, a routing
protocol that is suitable for data dissemination applications with less stringent delay constraints and
for use in mobile multihop wireless networks with intermittent connectivity. We first study the
inter-node contact behaviour and derive an inter-node contact model based on time series analysis.
Then, we present the details of ORION routing protocol by explaining the contact forecasting
process as well as the forwarding algorithm. In this contribution, a focus is given to resource
efficiency either in forecasting algorithm or queue occupancy for the store-and-carry-forward
mechanism.
Chapter 5: presents our third contribution; a stochastic topology control mechanism based
on wakeup/sleep states scheduling. For monitoring applications based on a static and dense
multihop wireless networks, keeping all the nodes active in the same time is not energy efficient.
7
First, and according to certain wakeup/sleep statistical distribution, we show that the best statistical
distribution parameters can be computed to meet certain requirements in terms of node's
neighbourhood size. After the estimation of these parameters, we propose distributed scheduling
algorithm that keeps a minimal topology working for achieving energy efficiency without
degrading the data delivery process both in terms of end-to-end delays and resource utilization.
Chapter 6: summarizes our contributions, and concludes this thesis. Future perspectives and
improvements to our contributions are also discussed.
8
Chapter 2
2 Challenged Multihop Wireless Networks
The recent advances in Micro-Electro-Mechanical Systems (MEMS) technologies [1] have
unlocked a wide range of potential applications [2] that make use of multihop wireless networks.
These networks are likely to be composed of hundred, and potentially thousands of small
wireless nodes powered on batteries, functioning autonomously, and often without access to a
renewable energy source. These nodes have the ability to sense the physical environment, process
the obtained information and communicate using short radio interfaces. Many applications
deployment fit to have a specific node called "sink node" which is responsible for collecting and
processing the information received from the other nodes.
Energy is considered as the main issue of these wireless networks. Indeed, such networks are
often considered for hostile environments, but also wide and wild areas in order to ensure specific
applications. Such applications include battlefield surveillance, volcano activity monitoring, wild
animals tracking, etc. Therefore, it is difficult to replace or recharge the battery. Thus, these
applications are deployed for just one lifecycle. Energy consumption has then to be reduced in
order to maximize the application lifetime and delay another deployment. Radio communication is
considered as the most energy consuming functionality. Since nodes cannot transmit data packets
over long distance to avoid draining energy, transmit power is adjusted to its minimum allowing
short range communications. This reality has led to a new communication paradigm: multihop
communications. In order to communicate data over long distances, wireless nodes act as relays for
other nodes in order to forward packets from their source towards the destination in a hop-by-hop
basis. Figure 1 represents an example of such networks.
Figure 1: Multihop Wireless Network Architecture
9
Furthermore, using a single path may quickly deplete the residual energy of individual nodes
across this path. This will result in unbalanced energy consumption across the sensing field, and
hence negatively affects the overall network lifetime.
Finding and establishing a routing path across the network may be very challenging if the
network topology has to become dynamic. Network topology is dynamic when using mobile
wireless nodes, but also in static networks where nodes adopt a wakeup/sleep scheduling in order
to reduce energy consumption, or simply due to nodes failure. In this context, the problem of
finding a routing path from the source to the destination evolves to a more challenging problem: of
finding a routing path in a changing topology in time.
Wireless node resources such as computing and memory capabilities are also an important
issue. Indeed, if these networks benefit from low manufacturing costs, they suffer from minimal
capabilities: less powerful CPUs, small memory size, etc. This fact makes complex algorithms
impossible to be implemented in such nodes, but advocates the design and implementation of
simple but efficient algorithms for all the functionalities: data gathering, data processing, packet
routing, topology control, etc.
Due to the characteristics of low cost and easy deployment of these networks, they have
attracted more and more attention and have been used in many different applications: Wireless
Sensor Networks (WSNs) [3][4][5], Wireless Multimedia Sensor Networks (WMSNs) [6], Wireless
Sensor and Actuator Networks (WSANs) [7], Wireless Mesh Networks (WMNs) [2], Underwater
Acoustic Sensor Networks (UASNs) [8], Mobile Wireless Sensor Networks (MWSNs) [9], Delay-
Tolerant Mobile Networks (DTMNs) [10], Unmanned Aerial Vehicles (UAVs) [11], etc.
2.1 Potential Applications Overview
Low cost wireless multihop networks, and especially WSNs, have been used for a variety of
applications such as military surveillance [11], environment monitoring [13], structural health
monitoring [14], asset tracking [15], domestic networks and home automation [16], smart places
[17], industrial process control [18], medical care [19], etc. According to [20], applications can be
categorized based on the data interaction patterns:
— Event-Detection: wireless nodes stay in a sensing state, and once they have detected a certain
event, they report this information the sink node. Sensing can be done locally and
continuously, but reporting can be triggered if certain rules apply such as a measured value
that exceeded a determined threshold. Some event-driven applications may require the
cooperation of a group of nodes in order to decide whether an event has occurred or no.
10
— Periodic Sampling: in such applications, nodes sense periodically the physical environment
and report the measurement to the sink node. Measurement can be triggered by the
application on a detected event (e.g. activating audio capture in a video surveillance
application when an intrusion has been detected). Another application based on periodic
measurement can be air pollution monitoring [21].
— Target Tracking: an event can be dynamic in time and space such as an intruder in the case
of surveillance applications. Wireless nodes can be used to measure and report to the sink
target information such as location, speed, direction, etc. Nodes have to cooperate between
them in order to consolidate an efficient report to be sent to the sink node.
— Actuators Control: since wireless nodes embed different types of hardware along the
physical sensors, such nodes can also be used in applications where an action can be needed
as a response to a certain detected event. In this cane, the node is not only part of the
identification process but it is part of the response process too. WSAN-based applications
fall into this category.
2.2 Wireless Node Architecture
Before describing the high level challenges that affect the multihop wireless networks, we
describe the node-level architecture for future references. Figure 2 represents a typical wireless node
with sensing capabilities:
Figure 2: Wireless Node Architecture
A wireless node consists of the following elements, some are essential whereas other
components may be optional:
— Sensing Unit, which includes one or several sensors along analog-to-digital (ADC)
converters (if necessary) for data collection.
11
— Processing Unit, which includes a microprocessor and memory in order to run any
processing code.
— Communications Unit, which includes a radio transmitter/receiver for wireless
communications.
— Power Supply Unit, which includes one or more batteries.
— Localization Hardware, any hardware that provisions the node with location
information such as GPS.
— Actuators: any hardware that is part of the application control system.
— Mobilizer: a hardware that allows the wireless node to move to another location
(motors, wheels, wings, …)
— Power Generator: any component to permit the recharging of the batteries or to be
used as primary energy source (solar panel, dynamo, …)
2.3 Low Cost Multihop Wireless Networks Characteristics
Unlike other wireless networks, such as ad-hoc networks, challenged multihop wireless
networks have the following characteristics:
— Application-specific. Low cost-multihop networks have gained a lot of attention and are
widely used for various applications as presented earlier. However, for different
applications, there are different needs and requirements. This fact has led to the apparition
of multiple network architectures and protocols.
— Easy deployment. Thanks to the distributed algorithms that allow an autonomous
behavior, these networks do not need any infrastructure establishment in advance. In
some applications, wireless nodes can be dropped for the air.
— Scalability. Due to the low manufacturing cost, these networks can be composed of
hundreds or thousands of nodes. Thus, deploying such networks allows the coverage of
wide areas.
— Changing Topology. Due to nodes' characteristics, connectivity may be highly dynamic
due to various reasons: mobility, short radio range, or nodes' failure. Network topology
dynamicity can also results from the deployment of new nodes in the same sensing field.
— Unattended Functioning. Such networks can be deployed in areas without the need for a
human attending such as hostile and dangerous environments (e.g. battlefield, volcano,
etc.). Thus, these networks must be self-organizing i.e. nodes organize themselves
automatically after deployment. Self-management and self-healing functionalities are also
required in order to deal with any network topology changes.
12
— Converged Traffic. Many applications based on such low-cost multihop wireless networks
rely on the presence of a special node (the sink node) which collects the data from the
other nodes. Multiple sink nodes may exist according to the application requirements.
The role of such node is the collection of the data but also the processing and the
generation of reports that describe the application functioning.
2.4 Challenges and Design Principles
The main factors that drive the architecture design of low cost multihop wireless networks
and especially WSNs [6] are:
— Fault-Tolerance,
— Scalability,
— Manufacturing cost,
— Hardware constraints,
— Physical environment,
— Network topology,
— Energy consumption.
These challenges are describes in the following sub-sections:
2.4.1 Fault-Tolerance
Fault –tolerance [22][23] is the ability to maintain the network functioning in the presence of
faults. The network reliability is affected by different faults due to different reasons such as node's
failure, energy depletion, bad deployment, etc. the failure of a single or a set of nodes should not
affect the overall functioning of the network. It is worth to notice that architecture and protocols
for these networks may be designed to address a predetermined level of fault-tolerance as perfect
fault-tolerance may be very difficult to achieve. This level may depend on the end-application. For
example, an application for weather monitoring may have a lesser fault-tolerance level than the
level for a battlefield surveillance application.
2.4.2 Scalability
Challenged multihop wireless networks may be composed of hundreds or, potentially,
thousands of nodes. This number may reach the extreme value of a million for some applications.
Designed algorithms and protocols must be capable of handling such number of nodes [24]. The
high density of these nodes within the sensing field must be considered too as it affects greatly most
of the nodes tasks (radio transmission, packet routing).
13
2.4.3 Manufacturing Cost
If the network is composed of a large number of nodes, the cost of a single node is very critic
as is affects the cost of the overall network [25]. If this cost is more expensive than the cost of
deploying traditional networks, then the use of this network is not cost-justified. Thus, the cost of
single node must be kept very low. Moreover, as nodes may embed other units, the cost of a single
node may become very challenging as it will consist of finding the best tradeoff between the
capabilities that may equip the wireless node and it total cost.
2.4.4 Hardware Constraints
The main hardware constraint is the size of the node [26] since all the utilized units must fit
into a small size module. Other constraints may be added such as the low energy consumption and
the resistance to extreme environment conditions (humidity, temperature, pressure, etc.). Overall,
since the network may be deployed for only one lifetime, the node hardware must resist and last as
long as possible.
2.4.5 Physical Environment
Nodes are often deployed within or very close to the studied phenomenon. Moreover, some
environments may be hazardous for human attendance. Thus, these challenged networks must
work unattended. Such environments include: biologically contaminated field [27], battlefield,
under water, volcanos, etc.
2.4.6 Network Topology
Deploying a large number of nodes requires efficient algorithms for topology maintenance
[28]. This topology maintenance can be composed of three stages: (1) deployment, (2) post-
deployment, and (3) redeployment. For deployment, nodes can be dropped in mass or placed one
by one in the sensing field. After deployment, changes to the topology must be handled. These
changes include node failure due to energy depletion or node destruction, nodes mobility, etc.
Finally, if the network topology cannot recover from severe topology changes such complete
partitioning, redeployment of new nodes may be necessary. These new nodes must then be placed
efficiently in order to recover the network topology from its partitioning.
2.4.7 Energy Consumption
Due to their small size, nodes are very limited in energy [29]. For some applications,
replacement or recharging of batteries is impossible. Since the node lifetime depends on the battery
lifetime, algorithms and protocols designed for these challenged networks must be energy-aware. In
some applications, where energy is not an issue (nodes placed in vehicles), the focus is on the
14
application requirements, such as quality of service (QoS), or resource efficiency in order to achieve
the desired application tasks with less powerful CPU and less memory.
2.5 Challenges across the Communication Protocols Stack
In this sub-section we present different challenges that researchers are facing at different
levels of the communications stack.
2.5.1 Applications Level Challenges
Applications challenges may include: quality of service, bandwidth, and data delivery model.
2.5.1.1 Quality of Service
In some applications, data must be delivered to the destination within a certain period of
time from the moment it has been sensed; otherwise the data will be useless. Therefore, end-to-end
delays should be bounded for time-constrained applications. However, in many applications,
conservation of energy, which directly impacts the network lifetime, is considered relatively more
important than the quality of the service. Hence, the design of time-constrained application using
low cost multihop wireless networks is based on a trade-off between the quality of service, and the
energy consumption.
2.5.1.2 Bandwidth
Some applications may require the sensing, the processing and the transmission of high
volumes of data as it is the case for multimedia applications. These applications face the problem of
limited link bandwidth. For example, it may be impossible to send a multimedia stream across a
single path in a hop by hop basis and to meet the delay constraints. To overcome this problem,
researchers propose the use of cooperative sensing where a group of nodes cooperate in order to
send partial information that is later reconstructed at the sink node. Another solution consists of
multiple path transmission, where the source node sends the sensed data across parallel disjoint
paths in order to boost the end-to-end transmission.
2.5.1.3 Data Delivery
Applications may have different requirements in terms of data delivery model [30].
— Periodic (or Time-driven). In this category, each node periodically sends data samples to
the destination node. Most monitoring applications rely on this data delivery model.
— Event-driven. In this category, nodes send data to the destination node once they detect
an event. Event-detection can be very simple such as monitoring a measured value until
it reaches a predetermined threshold, then a report is sent to the sink node. In some
15
applications, event-detection can be very complex such as in intrusion-detection
applications, where nodes cooperate among them in order to avoid the detection of
false positives, before reporting the event to the sink node.
— Query-driven. In this category, nodes send packets to the sink when they receive a
query packet from a control entity. For example, in medical care applications, the
doctor may query the data of one or some of his remote patients equipped with body
wireless sensor network.
— Hybrid. In this category, one or more of the previous data delivery models are used
simultaneously. For example, nodes can trigger a report to the sink upon the detection
of a certain event and periodic samples follow after for a certain period of time [31].
The data flow type impacts directly the application. For periodic data flows, compression or
in-network aggregation may be of good choice, in order to reduce the data packets that converge
towards the sink node. Other enhancement mechanisms may be envisioned too. For example,
avoid sending data samples that can be predicted by the sink node following a certain forecasting
algorithm. For query-driven applications, data may be made available at intermediate nodes such as
cluster heads, in order to enhance the query/response process. Thus, when a query is sent,
intermediate nodes can respond with the desired information. Finally, for event-driven model, a
minimal routing infrastructure can be kept in order to maximize the lifetime of the overall
network, since only few nodes can be used to relay data packets to the sink node.
2.5.2 Network Level Challenges
The main challenges at the network level are the dynamicity of the network topology and
topology voids.
2.5.2.1 Changing Topology
MWNs are often assumed to be stationary. However, nodes mobility is sometimes needed in
certain applications. Routing data packets may become very challenging since route stability is not
guaranteed. Changes to the topology can also be the result of node failure due to energy depletion
or physical damage. Energy conservation mechanisms such as wakeup/sleep scheduling may also
result in a dynamic network topology. Moreover, the studied phenomenon can be dynamic (for
example a moving target for a tracking application). In this case, the source node of the data is
dynamic. Changes to the network topology can be categorized according to the node mobility into:
— Stationary network: in these networks, node failure due to energy depletion or hardware
destruction is behind topology changes. The addition of new nodes is a source of
topology changes too. Duty-cycling can also be result in a dynamic topology.
16
— Mobile network: nodes mobility results in intermittent connectivity. Nodes face then the
appearing/disappearing of one-hop neighbours, making the forwarding decision very
difficult. For delay-tolerant applications, nodes can store the data packets until the
connexion with the destination node or a node that is certain or with high probability to
reach the destination node. In less tolerant applications, routing protocols should handle
this intermittent connectivity and include "navigation" functionality in order to let the
data packets find easily their path towards the destination node across mobile nodes or
messages ferries.
— Hybrid network: in hybrid networks, network topology dynamicity result from the
above factors (nodes mobility, duty-cycling and nodes' failure).
2.5.2.2 Topology voids
Voids are regions where no nodes have been deployed or a region where no nodes are longer
alive. Voids may also occur due to physical obstacles to radio communications. Handling topology
voids may be very challenging for packet routing. Indeed, nodes need to detect such voids and must
send data packets in a hop-by-hop basis in order to surround them. Topology control mechanisms
need to deal with these voids efficiently in order to avoid that data packets arrive to dead-end nodes
where no alternative forward path is available. Thus, routes heading to such topology voids need to
be pruned sufficiently in advance. Figure 3 present an example of nodes facing a topology void.
Figure 3: Presence of Void in the Sensing Field
2.5.3 Physical Level Challenges
MWN-based solutions are also facing various challenges at the physical layer such as link
synchronisation, transmission scheduling, and link control. However, challenges faced at this layer
of the communications protocol stack are out of the scope of this thesis and focus are given to high
level layers' challenges as described in the previous sections.
17
2.6 Analysis of Routing protocols for MWNs
In this sub-section we present a brief overview of the main routing protocols for MWNs and
especially WSNs [32][33] as we can find an impressive amount of routing protocols in the literature.
A simple taxonomy is represented in Figure 4.
Figure 4: Taxonomy of Routing Protocols for WSNs.
First, routing protocols can be organized according to the network structure that the end-
application is considering.
2.6.1 Network structure based routing protocols
The underlying networks structure can play an important role in the functioning of the
routing protocol. If all the nodes composing the network are homogenous in terms of tasks,
routing is considered flat. In the presence of node heterogeneity where nodes play different roles
and have different capabilities such as group managers or cluster heads, routing is considered
hierarchical. Finally when nodes rely on location information to perform packet forwarding
decision, routing is considered location-based.
2.6.1.1 Flat routing
Flat routing is a multihop routing where all the nodes are operational and are affected the
same tasks. In a flat topology, all the nodes cooperate in the data collection task. Due to the
deployment of a large number of nodes, it may be not possible to address the nodes individually
based on their identifier but regions of interest instead. This consideration is behind the data-centric
routing. Early works include Directed Diffusion [34] and SPIN [35] protocols. These two protocols
have influenced the design of many other protocols such as Rumor Routing [36], CADR [37],
COUGAR [38], ACQUIRE [39], etc.
18
2.6.1.2 Hierarchical routing
Flat architectures do not offer a great scalability. Indeed, as the number of deployed nodes
increases and the sensing field becomes wider, communications may become challenging. To
achieve scalability, hierarchical network architectures have been proposed. In these architectures,
nodes gather into groups or clusters and cluster-heads are elected. Special tasks are then affected to
these cluster-heads. This organization allows energy efficiency since it reduces communications
between cluster-nodes and other nodes of the network and data is collected, processed and
transmitted by the cluster-head. Hierarchical routing protocols include LEACH [40], PEGASIS
[41], TEEN [42], APTEEN [43], MECN [44], SAR [45], VGA [46], HPAR [47], TTDD [48], etc.
2.6.1.3 Location-based routing
In some applications, nodes are addressed using their location information. Thus, greedy
forwarding can be employed using distance or angle calculus. Location information can be obtained
from a specialized hardware such as GPS receiver or obtained using a distributed localization
technique. In distributed mechanisms, nodes rely on signal strength, angle of reception, etc. in
order to measure distances to landmark (that can be nodes with localization hardware). Such
routing protocols include: GPSR [49], GAF [50], GEAR [51], MFR, DIR, GEDIR [52], GOAFR
[53], SPAN [54], etc.
2.6.2 Protocol operation-based routing protocols
Based on the main operation of the protocol, routing protocols for WSNs can be organized
into the following categories:
2.6.2.1 Negotiation-based routing
Negotiation-based routing protocols use high level data descriptors in order to reduce
redundant data transmissions. SPIN [35] routing protocol falls into this category. The main idea
behind such protocols is pre-transmission stage where nodes negotiate the data transmission using
an advertisement/request mechanism. Thus, only interested nodes are requesting the data from the
source node.
2.6.2.2 Multipath-based routing
In order to enhance routing performance, many routing protocols rely on the use of multiple
paths. Multiple paths permit fault-tolerance since alternative paths exist in the case of the primary
path failure. Multiple paths allow boosting the end-to-end route capacity in terms of bandwidth
since parallel transmissions are possible. They also achieve lower end-to-end delays. However, the
construction and the maintenance of multiple paths may introduce extra overhead in terms of
control messages. Such protocols include TPGF [55], MPMPS [56], etc.
19
2.6.2.3 Query-based routing
In this case of routing protocols, the destination node (the node interested in certain
information) sends a request to a certain node or a group of nodes. Nodes that have the requested
data respond back with the desired information. These requests are often expressed in a high level
language. Directed Diffusion [34] protocol falls into this category.
2.6.2.4 QoS-based routing
In this kind of protocols, nodes are trading energy consumption for a better quality of service
in terms of lower end-to-end delays, transmission data-rate, etc. examples of such protocols include:
SAR [57], SPEED [58], etc.
2.6.2.5 Coherent-based routing
Coherent and non-coherent data processing differ mainly in the location of data processing.
In non-coherent data processing, collected data are treated locally at the source node before its
transmission to the sink node. However, in coherent data processing, collected data is sent directly
to an aggregator (i.e. a cluster-head) or to the sink node after minimum processing (i.e. value
redundancy elimination). Thus, routing protocols that use one of the previous data processing are
said coherent-based routing, or non-coherent-based routing [57].
2.7 Delay-Tolerant Routing in MWNs
In certain mobile deployment scenarios, connectivity can becomes very intermittent making
classical protocols unusable. Applications must tolerate large delays or the risk of no forwarding
guarantee. Such networks are called Delay Tolerant Networks (DTN). They mainly suffer from
intermittent connectivity [59] and make use of opportunistic communications [60]. Routing in such
networks is thus different from other wireless networks and is very challenging due to the frequent
partitioning of the network. For example, routing protocols for wireless networks such as AODV
[61] and DSR [62] may not be suitable for DTNs. Indeed, these protocols first establish a route
between the source node and the destination node, and then data transmission actually occurs. If an
instantaneous end-to-end path is hard to establish in DTNs, the store-carry-and-forward scheme
should be used instead.
DTNs have the characteristic that, unlike standard networks, an end-to-end path between the
source node and the destination node may not necessarily exist. However, it is desirable to allow
communications between nodes. In this context, classic routing protocol cannot deliver efficiently
data packets between the communicating nodes. Also, node mobility does not allow nodes to be
aware of the current locations of node, especially if the mobility pattern is completely unknown.
20
Delay-tolerant networks have been used in various situations such as deep space
communications, rural connectivity projects, city-wide ad-hoc networks, etc. DTN is also an
approach for hybrid networks convergence. Unlike the work presented in [63], the DTN
architecture presents an intermediate layer called the Bundle Layer that is located between the
transport and the application layer in order to achieve convergence between different protocols
stacks. The standard includes mechanisms for addressing, inter-node bundle transfer, security, etc.
The source of delay is mainly a result of 3 delays:
— Processing Delay: This is very short. It is the time necessary to process the packets
header at the routers.
— Queuing Delay: This delay may be the longest. It is the time spent in routers queues
waiting for transmission opportunity.
— Transmission Delay: This delay is wireless technology dependent. It is the time
necessary for transmitting a data packet through the wireless link. This delay also
includes the signal propagation delay which may be very significant. For example, it
takes about 14 minutes for a radio signal to be propagated from Earth to Mars.
In the following sub-sections, we give a brief overview about routing strategies proposed for
delay-tolerant networks. These strategies differ from each other in terms of the required knowledge
about the network in order to perform routing decisions.
2.7.1 DTN Characteristics
Delay-tolerant networks differ from conventional wireless networks. They exhibit the
following characteristics:
— Intermittent Connectivity [64]: DTNs networks suffer from network topology
petitioning due to nodes mobility. In some DTNs nodes are more likely to be
disconnected all the time and inter-nodes contacts occur very rarely. Moreover, these
contacts are prone to transmission errors, making efficient contacts that lead to actual
error-free data transmission rarer.
— Opportunistic Communications [65]: since connectivity is intermittent nodes achieves
data packet forwarding through opportunistic contacts. These contacts can be
scheduled and known in advance such as for deep space communications where the
nodes dynamics is well-known in advance (planet movements, satellites, space-craft,
etc.), or completely unknown and unpredictable such as moving people, vehicles, etc.
21
— Limited Longevity: Since DTNs networks can be deployed in much challenged
environments (space, oceans, deserts, etc.), and due to long delays, the overall
network lifetime can be very limited (energy depletion, attacks, physical damages,
etc.)
— Long Delays: inn DTN networks, an instantaneous end-to-end path may not
necessarily exist. Thus the delay that separate the data generation at the source node
and its reception at the destination node may be very long compared traditional
networks. Queuing delay in intermediate nodes may compose the big part of this
end-to-end delay.
— Low Data Rates: for some DTN scenarios, wireless links are used for very long
distances. Thus they are very limited in terms of data rate. Moreover, transmissions
errors may this data rate even lower. Also, such networks suffer from a low
instantaneous data rate since nodes are often disconnected and no data transmission
occurs.
— Packet Delivery Rate: as a result to the previous DTNs characteristics, packet delivery
rates are also very low. Indeed, packet can be sent on a hop-by-hop across a network
where a path may not exist towards the destination. Thus, packet loss is very
common in multihop DTNs. In order to maximize the delivery probability,
replication strategies have been widely used.
2.7.2 Routing Problem in DTNs
Most of routing protocols developed for delay-tolerant networks are designed using messages
replication. With such strategy, a message is replicated across contacted nodes in order to maximize
the probability of reaching the destination node. Functions are introduced to measure the
willingness of a node to deliver the message to its final destination and forwarding decisions are
made upon this willingness. Some functions need little or no knowledge about the network such as
Epidemic routing protocol [66] where every contacted node is a good candidate to deliver the
message and the routing is thus made through flooding.
To design an efficient routing protocol for delay-tolerant networks; some restrictions should
be taken into account:
— Resource Allocation [67]: in DTNs nodes are more likely to store relayed data packets
for long periods into their buffer queues, waiting for future contacts with other
nodes. This restriction need to be carefully addressed, since received packets after
queue saturation leads to their loss. Moreover, if message replication is considered,
22
resource may be then fewer across the networks, since copies of the same message
may occupy queues at different nodes.
— Buffer Space: as stated above, nodes need to have sufficient buffer space to host all the
potential data packets waiting for contact establishments due to intermittent
connectivity.
— Limited Energy: DTN nodes are often deployed unattended, i.e. nodes are battery-
powered. Communication task along the data processing and strong tasks may be
very energy-consuming (sending/receiving over long distances, algorithms for
contacts forecasting, maintaining large memory space, etc.). Thus routing protocol
need to be designed with the energy-awareness for all tasks.
— Contacts Availability [68]: since DTN nodes are often disconnected, communications
occurs through opportunistic contacts either scheduled/predictable or completely
unpredictable.
— Security Issues [69]: security may be a very serious problem for DTNs [70]. Since data
packets forwarding is delegated to nodes encountered in the network, sender and/or
message authenticity should be guaranteed. Secure end-to-end routing is desirable.
2.7.3 Routing Topologies for DTNs
An important issue for routing in delay-tolerant network is the stability of the routing paths.
The stability of a routing path may depend on various aspects such as the rate at which messages are
generated at the source. The more frequent is message generation, the most unstable the routes will
be, since it will be very difficult to find and maintain a route for a series of messages instead of a
route for a single message. Thus, this route stability depends on the topological changes based on
routing types:
— Unicast Routing: a source node will generate and send a single copy of the message. It
will be in charge of finding the best path to reach the destination relying on the
knowledge about the network topology.
— Broadcast Routing: a source generates a message and flooded it to all the nodes of the
network. Every flooded node carries a copy of the message. The message delivery
probability is very high in such routing.
— Location-based Routing [71]: data packets are forwarded greedily based on distance
calculus in order to reduce the distance that separates the source and the destination
node. However, there is no guarantee to reach the destination even if the distance is
relatively small (absence of appropriate relay nodes, nodes mobility, etc.)
23
— Tree-based Routing [72]: Tree-based routing can be seen as enhancement of broadcast
routing since it aims the reduction of the packet retransmissions. In this routing, the
source flood the message through a tree structure routed at the source reaches all the
receivers. Messages are replicated from parent node to child node.
— Cluster-based Routing [40][72]: nodes play different roles and special nodes are
responsible for forwarding messages within determined clusters in order to increase
message delivery.
2.7.4 Data Packet Forwarding Strategies
There are mainly two main categories of packet forwarding strategies: (1) forwarding-based
strategies, and (2) replication-based strategies [73]. Figure 5 presents a classification of routing
strategies for delay-tolerant networks.
2.7.4.1 Forwarding-based strategies
In forwarding-based protocols, only a single copy of the data packet is kept in the network
during its transmission from the initial source to the final destination. Data packets are transmitted
hop by hop through inter-node contact opportunities. These contacts depend on various factors
such as weather conditions, radio link interference, etc. Routing for the forwarded-based strategies
rely on the concept of custody-transfer where a node delegates the responsibility of the packet
forwarding to another relay node, until it will reach the destination node.
Figure 5: DTN Routing Strategies Classification
Forwarding-based strategies can be summarized into 3 classes:
24
2.7.4.1.1 Infrastructure-based strategies
In some applications scenarios, the mobility pattern of certain nodes may be known in
advance. Infrastructure-based strategies propose the deployment of a fixed mobile infrastructure
along the DTN nodes. Special mobile nodes have (data mules) have been used as messages ferries in
order to connect partitioned and disconnected parts of the network [74][75][76]. These mobile
nodes take messages from DTN nodes and move towards the destination node or the next relay
node, and deliver the picked carried messages. Such strategy has been shown to be very efficient
through extensive simulations in many research works. However, the deployment of an extra
fixed/mobile infrastructure cannot be considered in many real DTN scenarios.
2.7.4.1.2 Prediction-based strategies
Prediction-based strategies have been employed in many routing protocol for DTNs. Such
strategies aim to improve the routing performances through the calculation and the prediction of
the future network state (next contact date, delivery probability, next contact duration, etc.). A
typical routing protocol that uses prediction is PER [77]. PER predicts the message delivery based
on the probability distribution of future contact dates and chooses the appropriate next forwarder
aiming to enhance the packet delivery probability.
Prediction-based strategies are proven to behave better than traditional DTN routing
schemes in terms of delivery ratio and end-to-end delay.
2.7.4.1.3 Social-based forwarding strategies
Mobility patterns have been widely analyzed. Researchers have found that some mobility
patterns exhibits similar characteristics when compared to social networks. Thus, social
relationships have been extensively investigated. These social mobility characteristics are employed
in order to assist the routing decisions. Examples of such protocols that are based on social mobility
model are SimBet [78], SSAR [79]. SimBet routing protocol makes explicit use of complex network
analysis metrics and algorithms in order to highlight a node’s position in the aggregated social
graph, and assess its utility to act as a relay for messages destined to other nodes in the graph.
SimBet assumes that nodes naturally reside in mobility-related communities (e.g., class, work,
home). Thus, “well-connected” nodes in the network are chosen as messages ferries to relay
messages form one community to another, until a node that has many neighbors with the
destination, (i.e., belongs to the destination’s community) is reached.
In the other hand, SSAR routing protocol relies in the selfishness phenomenon of people for
forward selection in real life. SSAR model the network as weighted directed graph, where edge are
given a weight that corresponds to the willingness of the start node to forward packets to the end
25
node. These weights are real numbers in [0, 1]. A 0 value means that a node is unwilling to forward
packets to its connected neighbor whereas a 1 value means the connected neighbor will always be
chosen to forward messages. Edges' weights are set randomly at the initialization of the network
and then updated during the social changes.
2.7.4.2 Flooding or Replications-based strategies
Replications-based strategies require the relay nodes to keep a copy of the forwarded message
in order to increase the message delivery probability by increasing the number of potential nodes
that carry a copy of the message and that may encounter the destination node. This strategy can
enhance greatly the delivery rate and reduce the end-to-end delay. However, this enhancement is
made at the cost of network resources (memory, CPU, energy, etc.). Many research works have
been carried out in order to optimize the replication-based strategies in order to achieve reasonable
resource consumption. Epidemic Routing [66] is a well-known pure flooding-based routing
protocol for DTNs. With Epidemic routing, each node transmits the message to all the nodes it
encounters. Epidemic routing achieves a high delivery ratio reaching 100% if the storage space is
unlimited. However, it performs poorly in networks with limited resources.
2.7.4.2.1 Spray-based strategies
In order to limit the replication of messages across all encountered nodes as done in Epidemic
Routing, researcher studied and proposed protocols that limit the replication to some specific
nodes. These protocols operate according to two stages. First, multiple copies are created and
"sprayed" over the network. Then, each message copy is routed independently towards the
destination node. To limit the overall retransmissions in the network, the number of message
copies is relatively small and carefully decided. A well-known routing protocol that falls into this
category is "Spray-and-Wait" [80]. Spray-and-Wait protocol consists of transmitting L copies of the
originated messages to L different encountered nodes. These nodes will replicate the message to L
other distinct nodes, and so on. After this "spray" phase, nodes will wait for direct forwarding, i.e.
nodes will carry the message copy and wait until a direct contact with the destination node occurs.
An enhancement has been proposed in [81] namely "Spray-and-Focus". In the "Focus" phase, nodes
selects appropriates nodes based on utility function in order to forward a message copy to them.
Such protocols have been shown to achieve good routing performances in terms of end-to-end
latency and bandwidth overhead. Another enhancement s is also proposed in [82].
2.7.4.2.2 Social-based replication strategies
As social-based forwarding strategies, social-based replication strategies make used of social
network models along message replication to make routing decisions in order to increase de the
26
message delivery rate and end-to-end delays. Such protocols include BUBBLE [83] and SocialCast
[84]. BUBBLE uses the same social network model as SimBet. The data packet forwarding is
divided into global forward and inter-community forward. SocialCast is a routing protocol which is
based on the publish-subscribe concept. It relies on metrics of social interactions about nodes in
order to determine the best relay node.
2.7.4.2.3 Intention-oriented strategies
Some proposed routing protocols for delay-tolerant networks treat routing decision as an
optimization problem. RAPID [85] routing protocol falls into this category. RAPID models the
data packet routing as a resource allocation problem, and expresses routing metrics as per-packet
utility in order to determine how packets should be replicated in the system. For example, in order
to minimize the average delay, the utility of a packet i can be expressed as ( ), where
D is the accumulated delay. Replicating the packet to a certain node is then subject to the
minimization of the accumulated delay so far and the increment in the delay if the message is sent
through this encountered node. In [86] authors propose an energy-efficient forwarding algorithm
based on epidemic routing. In this protocol, each message has its own energy constraint that is
proportional to the number of expected transmissions in its life span in the network.
2.7.4.2.4 Coding-based strategies
In [87][88], data packets are fragmented and network coding is used in order to reduce
resources consumption. In such protocols, data packets are fragmented into multiple fragments at
the source node. These fragments are then replicated and broadcasted in the network. Intermediate
nodes combine these fragments, encode them again and a new packet is retransmitted. Depending
on the encoding algorithm, if the packet was initially encoder into k fragments, then the destination
node can attempt to decode the original packet after receiving k different fragments. Though this
method can use low buffer space, the routing may suffer from a longer delay if the k fragments fail
to arrive at the destination node. However, coding-based strategies are proven to achieve good
routing performances in the case of limited bandwidth and buffer space.
2.7.4.3 Hybrid-based strategies
Some research works have proposed the use of both forwarding and replication strategies
simultaneously in one routing protocol. Such protocols aim to find the best trade-off between the
high delivery rate brought by replication mechanisms and the resource efficiency brought by
forwarding mechanisms. Such protocols include Max-Contribution [89] and HYMAD [90].
27
2.8 Self-Organization in MWNs
In certain scenarios, the physical deployed topology is not suitable for the well-functioning
application. This underlying topology may be very resource inefficient in term of energy
consumption. For example, in densely deployed networks, nodes may have a very large number of
one-hop neighbors leading to radio transmission issues (interference, delays due to back-off
algorithms) but also energy depletion due to intense radio activity.
Self-organization is the process through which nodes organize themselves autonomously into
a logical network topology more efficient than the physical underlying topology. Self-organization
aims to provide efficient, self-adaptive, scalable, fault-tolerant and robust communications protocols
for dynamic and distributed multihop wireless networks.
Self-organization techniques can be classified into two main classes: (1) link pruning
techniques where all the nodes are part of the logical topology and some links are "disconnected",
and (2) dominating set techniques where only subset of nodes are composing the logical topology
while the other nodes can be turned off.
2.8.1 Link pruning techniques
Due to the lack of centralized infrastructure in wireless multihop networks, and the dynamic
topology, a fixed topology is not possible. The main goal of a topology control technique is then to
construct an appropriate topology in order to overcome these issues. Link pruning algorithms rely
heavily on geographic locations of nodes or on accurate distances between the nodes. In the
following we present the main link pruning techniques self-organization.
2.8.1.1 Relative Neighborhood Graph
The Relative Neighborhood Graph (RNG) [91] H of a graph ( ) (Where V denotes
the vertices set, and E the set of edges) is defined by ( ) where E' is the set of edges that
obey to the following rule: "An edge that links two vertices u and v is part of H if and only if it
does not exist any vertex w closer to both u and v". This rule can be expressed as follows:
( ) ( )
Where represents the Euclidian distance between the two vertices u and v. Thus, the
RGN graph removes the longest edge in the triangle uvw (as shown in Figure 6).
RNG is a fully local algorithm since it only requires the knowledge of the one-hop
neighborhood in order to construct the logical topology. However, the RNG algorithm relies on
the knowledge of nodes locations or the distance between them, as well as the radio transmission
28
range. In [92] authors have shown that the minimum spanning tree (MST) is a sub-graph of the
RNG. Thus, the RNG possesses the connectivity property of the initial graph (see Figure 12).
RNG has also the characteristic to make two neighboring nodes in the physical topology very
distant in the logical one. The average node degree in the RNG is close to 2.5 [93].
Link pruning algorithms are often used in geographical routing protocols [94]. They offer a
planar graph which aids the selection of the next forwarder node. RNG is used in [95] in order to
reduce the number of retransmission when flooding data packets in the network.
Figure 6: Link Pruning Mechanism in RNG Figure 7: Link Pruning Mechanism in GG
2.8.1.2 Gabriel Graph
If we consider the disc ( ) with diameter , an edge is part of the Gabriel Graph (GG)
[96] if and only if ( ) does not include any other vertex (cf. Figure 7). We can notice that RNG
is a sub-graph of GG. Thus, GG preserves the connectivity property of the initial graph. A GG
graph of 100 nodes is shown in Figure 11.
2.8.1.3 Delaunay Graph
Delaunay Graph (DG) is defined as the dual of the Voronoï diagram. A Voronoï diagram is
the union of multiple Voronoï regions. A Voronoï region is defined as the region including all the
points that are closer to a certain node x than any other node. Node x is implicitly put at the center
of this region. From a Voronoï graph and for every pair of nodes whom Voronoï regions are
adjacent, an edge is created between these two nodes in the Delaunay graph. Figure 8 shows an
example of a Delaunay graph along a Voronoï diagram.
29
Figure 8: Delaunay Graph along Voronoï Diagram
2.8.1.4 Local Minimum Spanning Tree
A spanning tree is sub-set of a non-oriented convex graph that is a tree and does connect all
the vertices. If a cost is assigned to each edge in the graph, the cost of the graph is then the sum of
the edges costs. A spanning tree is said minimum if it has the smaller cost among the entire
spanning trees of the graph. In this context, the cost of an edge is the Euclidian distance between
the two nodes connected through it. With the Local Minimum Spanning Tree [97], every node
computes its neighborhood minimum spanning tree (MST) using algorithms such as Prim
Algorithm [98]. The LMST construction is governed by the following rule: "an edge between two
nodes u and v is part of the LMST if and only if v is a neighbor of u in the MST of u, and if v is a
neighbor of u in the MST of v (as shown in Figure 9). Finally the LMST topology preserves the
connectivity of the initial graph (see Figure 13).
Figure 9: Link Pruning Mechanism in LMST.
30
It is worth to notice that LMST topology is achieved without global knowledge of the
network. However, the LMST algorithm relies on accurate distance measuring between the nodes.
Many routing protocols rely on these link pruning techniques in order to guarantee the packet
delivery by using a planar network topology.
2.8.2 Dominating Set Techniques
A Dominating Set (DS) of a graph is defined as a subset of nodes where each node is either
part of this subset or it is a neighbor of a node that is part of this subset. This subset is called
Connected Dominating Set (CDS) is the sub-graph based upon this subset is connected. CDS
algorithms allow the reduction of hop count between the source and the destination. Constructing
CDSs with minimum cardinality is NP-Hard problem. Thus, many heuristics have need proposed
to compute such subset. An example of a connected dominating set according to [99] is shown in
Figure 15.
The main Dominating Set techniques are:
2.8.2.1 Multipoint Relay and Multipoint Relay Dominating Set
Multipoint Relay (MPR) has been proposed in [100] and it is behind the standardized routing
protocol Optimized Link State Routing (OLSR) [101]. Authors used the resulting dominating set in
order to reduce the number of packet retransmission in flooding operations. In order to construct
the MPR nodes subset, each node selects among its neighbors, the nodes that allow reaching all 2-
hop neighbors. These neighbors are called MPR nodes.
In order to build a topology that is connected and dominating, every MPR node retransmits
a topology construction control packet while pointing out its MPR nodes in its local
neighborhood. Thus, the designation of MPR nodes is localized. However, the construction of the
dominating set is distributed across the network and depends on the considered source node.
MPR-Dominating Set (MPR-DS) [102] uses the same relay selection as in MPR. However, in
order to avoid the dependence toward a source node that initiates the topology construction, it is
based upon nodes identifiers leading to a totally distributed algorithm independent from the source
node. A node is part of the dominating set if: (1) its identifier is the smallest among its direct
neighbors, or (2) it has been designated MPR node by the neighboring node with the smallest
identifier.
31
Figure 10: Initial Unit Disc Graph[103] Figure 11: Gabriel Graph (GG) [103]
Figure 12: Relative Neighbour Graph (RNG) [103] Figure 13: Local Minimum Spanning Tree (LMST) [103]
2.8.2.2 Connected Dominating Set – Rule k
Dai and Wu [99] proposed a localized construction of the CDS. This proposed algorithm
operates through two stages. The first stage concerns the marking of the nodes, while the second
stage concerns link pruning according to a rule named "rule k". Basically, a coverage of a node can
be withdrawn of its neighbor set can be collectively covered by those of k coverage nodes.
32
These k coverage nodes have higher priority and are connected. Moreover, the CDS derived
from the marking process with Rule k can be locally maintained, when sensors switch-on/off.
2.8.2.3 Connected Dominating Set – Independent Dominating Set
Authors in [103] propose the construction of a CDS-ID. A dominating set is called
independent if each node part of this subset is not adjacent to another node that is also part of this
subset. Thus, nodes that are part of IDS are exactly 2-hops away from each other. This algorithm
operates through 3 steps: (1) construction of the spanning tree, (2) construction of the IDS, and (3)
construction of the dominating tree. The proposed algorithm is totally distributed. Since, the
dominating set is constructed after a network discovery phase; this algorithm achieves a better
selection of the nodes to be part of the connected dominating set.
Figure 14: Initial Unit Disc Graph[103] Figure 15: Connected Dominating Set[103]
Self-organization techniques which are based on either connected dominating sets or link
pruning, aim to overcome the inefficient physical topology resulting from network deployment.
These techniques build a logical topology upon the physical one in order to reduce the energy
consumption.
2.9 Conclusion
In this chapter, we have presented a brief state of the art of challenged multihop wireless
networks. We first, gave an overview of the potential application that can be considered using these
networks.
33
After a brief description of the architecture of a wireless node, we presented the different
characteristics of these networks. These networks are generally application-specific and easy to
deploy. They are self-organizing and function unattended. Their network topology is often
dynamically changing, and exhibit traffic convergence towards a central entity called the sink node.
In the second part, we presented the general challenges and design principles that application
design should consider carefully. Thus application developed for such networks should be fault-
tolerant, scalable, and energy-aware. Also, at the node level, hardware constraints and
manufacturing cost should be taken into account. After that, we present a simple taxonomy of
some challenges according to their level in the communication stack, mainly application specific
challenges and network specific challenges. The first class includes challenges related to the quality
of service, bandwidth and the data delivery model, while the second class includes challenges related
to the changing network topology and topology holes.
Then, we present taxonomy of routing protocols for WSNs as an example of challenged
MWNs where energy utilization is at the heart of the application functionalities design and
especially the routing functionality. For QoS-based applications, multipath routing can be of good
choice. However, most of the surveyed protocols rely on important control traffic for the topology
discovery and the routing path construction and maintenance. This overhead may lead to routing
inefficiency.
DTNs are another example of challenged MWNs, where delay is at the centre of the
application design. We start by presenting these networks and their characteristics, then we present
taxonomy of routing protocols proposed for DTNs. However, a lot of these protocols rely on
replication-based schemes or extended knowledge about the network topology in order to perform
routing decisions.
Finally, multihop wireless network self-organization is presented and state-of-the art
techniques are reviewed. These techniques agree to construct an efficient logical topology upon the
physical network.
34
Chapter 3
3 Online Multipath Routing in Multi-hop Wireless Networks
3.1 Introduction
With the advancement in miniaturization and the availability of low-cost hardware, the
computing nodes embed various kinds of sensing and capturing elements including microphones
and video cameras. Hence, the use of ubiquitous Wireless Multimedia Sensor Networks (WMSNs)
is becoming a reality [4][105][106][107].
WMSNs are generally used for surveillance applications, intrusion detection, environmental
and building monitoring, etc. These applications imposes additional challenges such as energy-
efficient data processing both within node and in-network, audio/video bandwidth/rate adaptation
to overcome the variations in networking conditions, Quality of Service (QoS) delivery to meet
application specific requirements and routing and selecting appropriate paths for continual delivery
of multimedia streams. Due to the distributed and dynamic nature of these types of networks, the
design of a critical information infrastructure based on a WMSN raises many other challenges such
as ensuring confidentiality and the integrity of the data stream, providing the means for node
authentication and access control, securing routing and node grouping [108]. Among all these
challenges, our work focuses on the routing and path selection issues taking into account energy
constraints and QoS delivery needs.
Generally, routing in wireless sensor networks (WSN) is a challenging task. A comprehensive
survey of routing protocols in WSN is given in [105]. A large number of research works exists to
enable energy efficient routing in WSN. In fact, we can find different routing techniques that try to
achieve energy-efficiency and to provide a best quality of service. One example is the multi-channel
transmission in WMSNs. In [109], authors have evaluated the performances of routing (routing
delays) when using a single and multi-channel communications in a wireless sensor and actor
networks. The authors showed that the multi-channel scheme performs better than the single
channel scheme especially for higher volumes of generated traffic putting the light on the important
need to parallel transmissions in a wireless multimedia sensor network, where delay and packets
loss are stringent constraints. Even in IP networks, multimedia transmission is still challenging.
Although the use of traffic prioritization in DiffServ networks, end-to-end throughput is not
guaranteed, application can make use of dynamic content adaptation based on end-to-end path
35
measurement [110][111]. In wireless IP networks, when data transmissions are prone to errors,
application can make use forward error-correction mechanisms in order to cope with random
packet loss [112][113].
In higher layers of the communication protocols stack, performances evaluations of routing
protocols for WMSNs suggests multipath routing approach to maximize the throughput of
streaming multimedia traffic. This is to utilize diverse paths to route packet streams towards the
destinations in order to avoid draining the energy of nodes along a specific route. In [114], the
authors propose a multipath routing protocol based on the well-known routing protocol Directed
Diffusion [33] that reinforces multiple routes with high link quality and low latency. In [115], the
authors focused on two key questions regarding multipath routing in WMSNs: (a) how many paths
are needed? And (b), how to select these paths? The authors then proposed a multipath routing
mechanism in order to provide a reliable transmission environment with low energy consumption
by utilizing the energy availability and the received signal strength of the nodes to identify multiple
routes from the source to the destination. In [116], the author addresses the problem of interfering
paths in a WMSN and considers both intra-session as well as inter-session interferences. The author
proposes an incremental path creation mechanism where additional paths are set up only when
required (typically in case of congestion or bandwidth shortage). In [117], authors propose MCMP
(MultiConstrained MultiPath) routing protocol in order to guarantee a better QoS in terms of
delay and reliability. Unlike end-to-end QoS schemes used in WSNs, the authors utilize a multiple
paths creation mechanism based on local link information.
Other examples of multipath routing protocols for WMSNs include: MPMPS (Multi-Priority
Multi-Path Selection) [56] and TPGF (Two-Phase Geographical Greedy Forwarding) [55]. However,
these “offline multipath” protocols have to explore the multiple routes that may exist between the
source and the destination before the actual data delivery phase. They may not be well adapted for
large-scale highly dense network deployments and for networks with frequent node mobility.
Geographic routing is the process in which each node is aware of its geographic coordinates
and uses the position of packet’s destination to perform routing decisions. These types of routing
scales better for WSNs. Greedy Perimeter Stateless Routing (GPSR) [49] was defined as a
geographic routing protocol in order for the network to scale in large size networks, i.e., to
accommodate a large number of nodes having very low exchange of route state information and
maintenance. The advantage of this protocol is that each node only gathers the topology
information about its immediate neighbours. Thus, its greedy forwarding relies on local-knowledge
for selecting the closest next hop node to the destination. This process ends up with continuous
36
selection of the same path that leads to fast depletion of the energy of the nodes along the selected
route and premature dying of these nodes.
In this chapter, we examine the benefit of geographic routing along with “online” multi-path
route selection process (i.e. multiple routes are created as packets advance towards the destination)
and propose a new routing protocol called AGEM (Adaptive Greedy-compass Energy-aware
Multipath) that takes into account both node’s energy constraints and QoS needs of audio and
video streams.
The design of AGEM is driven by the following factors:
Alternative paths: multimedia applications are delay sensitive and have delay and delay
variation constraints. Multimedia traffic should be delivered satisfying these requirements. In
typical networks, shortest paths are heavily used for the delivery of this traffic types whereas other
appropriate routers that could satisfy these traffic requirements are under-utilized.
Load balancing: In order to maximize the lifetime of WSN nodes and to avoid depletion of
nodes’ energy and consequently node’s failures, load balancing and multi-path delivery across the
network must be considered during the design of a routing protocol.
Multipath transmission: Packets in a multimedia stream are generally large in size and the
transmission requirements can be several times higher than the maximum transmission capacity of
sensor nodes if a single path is used for routing these packets.
Online decisions: As the topology may change from time to time, it is more appropriate to
make the routing decisions in a distributed manner and in real-time. This is due to the fact that
offline routing processes cannot react to topology changes and result in forwarding packets to
unavailable nodes or towards disconnected routes.
Node selection process: in densely deployed networks, different neighbours may be selected as
candidate for packet forwarding. To deduce an appropriate selection, the node selection process
should take into account, node’s energy, its distance to the destination, and packet’s QoS
requirements.
The rest of this chapter is organized as follows. Section 3.2 reviews the related work in the
area of WSN routing that influenced the design of our proposed protocol. Section 3.3 presents the
functionalities of proposed AGEM protocol. Section 3.4 provides the results of performance
evaluations of our proposed protocol in comparison with GPSR. Finally, section 3.5 presents our
conclusions.
37
3.2 Related Work
In geographic routing, two greedy schemes are used to make packets progress towards the
destination node. Greedy progression scheme based on distance to the destination node [49][118]
[119][120] and greedy progression based on angular offset in the direction towards the destination
node [121][122][123]. In both schemes, a route between source and destination is progressively
chosen only based on node-level forwarding decisions made locally at each hop.
For WMSNs, two important protocols have been proposed that make use of node positions
for packet forwarding i.e., GPSR and MPMPS. MPMPS is itself based on TPGF. These protocols
are further explained below.
3.2.1 The GPSR Routing Protocol
The GPSR (Greedy Perimeter Stateless Routing) [49] was originally designed for MANETs but
rapidly adapted for WSNs. The GPSR algorithm relies on the correspondence between the
geographic location of nodes and the connectivity within the network by using the location
position of nodes to forward a packet. Given the geographic coordinates of the destination node,
the GPSR algorithm forwards a packet to destination using only one single hop location
information. It assumes that each node knows its geographic location and geographic information
about its direct neighbours.
GPSR protocol uses two different packet forwarding strategies: Greedy Forwarding and
Perimeter Forwarding. When a node receives a packet destined to a certain node, it chooses the
closest neighbour out-of itself to that destination and forwards the packet to that node. This step is
called the Greedy Forwarding. In case that such node cannot be found, (i.e. the node itself is the
closest node to the destination out-of its neighbours but the destination cannot be reached by one
hop), the Perimeter Forwarding will be used. The Perimeter Forwarding occurs when there is no
neighbour closest to Destination (D) than node (A) itself. Figure 16 illustrates that node A is closer
to D than its neighbours x and y. This situation is called “voids” or holes. Voids can occur due to
random nodes deployment or the presence of obstacles that obstruct radio signals. To overcome
this problem, Perimeter Forwarding is used to route packets around voids. Packets will move
around the void until arriving to a node closest to the destination than the node which initiated the
Perimeter Forwarding, after which the Greedy Forwarding takes over.
38
D
Ax
y
xD
AD
yD
AD < xD
AD < yD
Figure 16: GPSR Perimeter Forwarding to Bypass a Void.
By maintaining only information on the local topology, the GPSR protocol can be suitable
for WSNs. However, the greedy forwarding leads to choose only one path from the source to the
destination.
3.2.2 The TPGF Routing Protocol
TPGF (Two Phase geographical Greedy Forwarding) [55] routing protocol is the first to
introduce multipath concept in wireless multimedia sensor networks (WMSNs) field. This
algorithm focuses in exploring and establishing the maximum number of disjoint paths to the
destination in terms of minimization of the path length, the end-to-end transmission delay and the
energy consumption of the nodes. The first phase of the algorithm explores the possible paths to
the destination. A path to a destination is investigated by labelling neighbours nodes until the base
station. During this phase, a step back and mark is used to bypass voids and loops until successfully
a sensor node finds a next-hop node which has a routing path to the base station. The second phase
is responsible for optimizing the discovered routing paths with the shortest transmission distance
(i.e. choosing a path with least number of hops to reach the destination). The TPGF algorithm can
be executed repeatedly to look for multiple node disjoint-paths. It’s worth to note that TPGF is an
offline multipath routing protocol.
3.2.3 The MPMPS Routing Protocol
The MPMPS (Multi-Priority Multi-Path Selection) [56] protocol is an extension of TPGF.
MPMPS highlights the fact that not every path found by TPGF can be used for transmitting video
because a long routing path with long end-to-end transmission delay may not be suitable for
audio/video streaming. Furthermore, because in different applications, audio and video streams
play different roles and the importance level may be different, it is better to split the video stream
into two streams (video/image and audio). For example, video stream is more important than audio
stream in fire detection because the image reflects the event, audio stream is more important in
deep ocean monitoring, while image stream during the day time and audio stream during the night
39
time for desert monitoring. Therefore, we can give more priority to the important stream
depending on the final application to guarantee the using of the suitable paths.
3.2.4 Policies for Greedy forwarding
In literature, there are different policies that can be used in geographic routing and for the
selection of the next hop node. To illustrate these policies, let take ‘u’ as the current forwarder
node and ‘d’ the destination node, then we can define these routing policies (see Figure 17):
Compass routing: See Figure 17(a) – The next relay node is ‘ ’ such that the angle is
the smallest among all neighbours of ‘ ’ [121].
Random compass routing: See Figure 17(b) – Let ‘ ’ be the node above line ( ) such that
is the smallest among all such neighbors of ‘ ’. Similarly, define ‘ ’ to be node below line
( ) that minimize the angle . Then, node ‘ ’ randomly chooses ‘ ’ or ‘ ’ to forward the
packet [121].
Greedy routing: See Figure 17(c) – The next relay node is ‘ ’ such that the distance ‖ ‖ is
the smallest among all neighbours of ‘ ’ [124].
Most forwarding routing (MFR): See Figure 17(d) – The next relay node is ‘ ’ such that
‖ ‖ is the smallest among all neighbours of ‘ ’, where ‘ '’ is the projection of ‘ ’ on segment
[124].
Nearest neighbour routing (NN): See Figure 17(e) – Given a parameter angle ‘ ’, node ‘ ’
finds the nearest node ‘ ’ as forwarding node among all neighbours of ‘ ’ in a given topology such
that .
Farthest neighbour routing (FN): See Figure 17(f) – Given a parameter angle ‘ ’, node ‘ ’
finds the farthest node ‘ ’ as forwarding node among all neighbours of ‘ ’ in a given topology such
that .
Greedy compass: Node ‘ ’ first finds the neighbours ‘ ’ and ‘ ’ such that ‘ ’ forms the
smallest counterclockwise angle and ‘ ’ forms the smallest clockwise angle among
all neighbors of ‘ ’ with the segment . The packet is forwarded to the node of with
minimum distance to ‘ ’ [125][122].
40
(a)
(b)
(c)
(d)
(e)
(f)
Figure 17: Greedy Forwarding Strategies: (a) Compass Routing; (b) Random Compass Routing; (c) Greedy
Routing; (d) Most Forwarding; (e) Nearest Neighbour Routing; (f) Furthest Neighbour Routing.
3.2.5 Discussion on Routing/Forwarding
Paths are selected a priori by protocols such as TPGF and MPMPS. In such cases, paths are
chosen in advance from the source to the destination. Knowing the full map of the deployed
network to perform routing as done by most “offline multipath” routing protocols is not suitable
for many reasons: (1) the exchange of the network map is energy consuming, (2) the map may not
reflect the current network topology, and (3) nodes’ failure can be more frequent in WSN than in
other ad-hoc networks. These reasons cause routing problems. In GPSR protocol, packets are
forwarded hop by hop based on information available local to node i.e., the use of “Greedy routing”
policy. GPSR seems to be more promising to scale to large network but does not achieve load
balancing by making use of multiple routes.
Hence, we propose a new geographical and online routing protocol called AGEM that (1)
selects neighbour nodes using an adaptive compass mechanism which is a newly defined policy, (2)
routes packets on multiple paths using greedy routing policy for load balancing purposes, and (3)
avoids network holes using walking back forwarding.
3.3 AGEM Routing Protocol
The main idea behind AGEM[126][127] protocol is to include a load-balancing feature while
being a greedy geographic routing protocol in order to increase the lifetime of the network and to
reduce the queue size in the most used nodes across the network. While using a pure greedy routing
protocol like GPSR, data/video streams always use the same route. In AGEM routing protocol,
data/video streams are routed using different paths. At each hop, a forwarder node decides to
41
which neighbour to send the packet. The forwarding policy at each node is based on the following
four parameters: (1) the residual energy at node, (2) the number of hops visited by the packet before
it arrives at this node, (3) the distance between the node and its neighbours, and (4) the history of
the packets forwarded belonging to the same stream. Furthermore, only a subset of available
neighbours is chosen according to the new adaptive compass selection mechanism.
The AGEM routing protocol has two modes, the Smart Greedy Forwarding and the Walking
Back Forwarding. The first mode is used when there is always a neighbour node closer to the
destination node than the forwarder node. The second mode is used to get out of a blocking
situation in which the forwarder node can no longer forward the packet towards the destination
node. Figure 18 presents an overview diagram of AGEM routing mode switching.
The following section will explain the two routing modes.
Walking Back
Mode
Smart Greedy
Mode
CN : Number of
Closer Neighbors
to the Sink
CN > 0
CN = 0
Figure 18: GEAMS Routing Mode Switching.
3.3.1 Smart Greedy forwarding mode:
AGEM is a geographic routing protocol where the nodes are aware of their geographic
coordinates. This information can be obtained using a positioning system such as GPS or by using
distributed localization techniques such as DV-Hop [128], Amorphous [129], etc.
In AGEM routing protocol, each sensor node keeps track of related information about its
immediate neighbours and stores the information that includes the estimated distance to its
neighbours, the distance of the neighbour to the destination, the data-rate of the links, and the
remaining energy of neighbours. This information is updated by the mean of beacon messages
propagated locally, scheduled at fixed adjustable intervals. Relying on this information, a forwarder
node will give a score to each neighbour according to a function (i.e. “f(x)”).
Since AGEM protocol is an online protocol and relies on beacon exchange for
neighbourhood state maintenance, AGEM can be used for static or mobile sensor networks.
Since AGEM routing algorithm is based on geographic coordinates, distance-based greedy
progression is used along angle-based greedy progression for next hop node selection. So, not all the
42
neighbours closest to the destination than the forwarder node are going to be selected as the
candidates for packet forwarding. This set of nodes is reduced to only include those nodes with best
angular offset towards the destination.
At the beginning, the forwarder node chooses only neighbour nodes that are within an
angular (α) view towards the destination with an initial angle of α0 (e.g., α = α0 <30°). A minimum
of “n” neighbour nodes (neighbouring set with n>=2) must be found to perform load balancing. If
n=1 then there is just one node set where no load balancing can be achieved. If no node is found,
the angle α is incremented by Δα (e.g., Δα =10°) until it reaches 180°. At this stage, if no node is
found then a walking back forwarding is needed since the forwarder is facing a hole. Figure 19
illustrates this adaptive forwarding policy. The angle of view is chosen small and incremented until
finding the desired number of candidates' nodes. This way, the selected nodes will always have the
least angular offset from the line (src)→(dst). Moreover, these selected nodes are then ordered
according to their given score.
α = 30°α = 60°α = 9
0°
α =
180
°
SinkForwarder
Node
Figure 19: AGEM Adaptive Compass Policy.
Choosing a node from the neighbouring set to forward a packet will depend on the score
given to each node according to the “f(x)” function (see Figure 20). The f(x) considers the energy
consumption which is defined in the following subsection.
43
SinkSource
one-hop
Neighbors
Forwarder
Node
Neighbor N1
Neighbor N2
Neighbor Nm
Figure 20: One-hop Neighbours Sorted According to their Scores.
3.3.1.1 Packet energy consumption :
When a node ( ) sends a packet ( ) of bits size to a node ( ), the energy of node ( ) will
decrease by ( ) while the energy of the node will decrease by ( ). Consequently, the
cost of this routing decision is ( ) ( ) considering the energy of the whole network.
Figure 21 illustrates this energy consumption.
Node A Node Bpk
n Bits
– ERX ( n )– ETX ( n, AB )
Figure 21: Packet Energy Consumption between two Communicating Nodes A and B.
We assume that the transmitted data packets in the network have the same size. We propose
an objective function to evaluate a neighbour for packet forwarding. This objective function
takes into account the packet energy consumption and also the initial energy of that neighbour.
The proposed objective function can simply be:
( ) (
)
Where: ( ) is the estimated energy to transmit a data packet through a distance D, and
is the estimated energy to receive the data packet.
These two functions rely on the energy consumption model proposed by Heinzelman et al
[130]. According to this model, we have:
( ) ( )
( )
44
Where: is the size of the data packet in bits, is the transmission distance in meters,
is the energy consumed by the transceiver electronics, is the energy consumed by the
transmitter amplifier. was taken to be and .
Upon receiving a data packet from the source node , the forwarder node retransmits the
packet to a neighbour that is closest to the destination node and in such a way that the number of
hops the packet traversed, will meet the rank of that neighbour (neighbours are ranked according
to their score). The main idea is to forward a packet with the biggest number of hops through the
best neighbour, and consequently a packet with the smallest number of hops is routed through the
worst neighbour to allow a proper load balancing in the network (see Figure 23 and Figure 24).
Figure 22 describes an algorithm as the forwarding policy.
For each known source node a forwarder node (N) maintains a pair ( ). represents
the mean hop count that separates from N, and j represents the neighbour (Nj) whom score (i.e.
f(x) function) is closest to the average score of all closest nodes to the sink in the neighbour set
(called best neighbour set).
As shown in Figure 22, the algorithm checks (Line 1) if a packet is already received from a
source node. If no, the packet will be always forwarded to the best node (line 2), and the hop count
“H” and the average score index “j” in the best neighbour set are set. These empirical values will be
used later to allow load balancing. It is clear that the first packet received from an unknown source
will be always forwarded to the best neighbour node.
Upon_Recieving_a_Packet ( pk )
Parameters:
Best_Neighbor: a set of the closest neighbours to the sink node sorted in descending order by
their score {BN1, BN2, … BNm}.
m = |Best_Neighbor|. m represents the cardinal of the Best_Neighbor set
j :index of the node in the set Best_Neighbor whom score is closest to the average score of all
closest nodes to the sink. For example, if Best_Neighbor is {8,5,2,1} the average score is 4 then
j=2 (starting from index=1)
Functions:
Get_Hop_Values (Si) returns the stored values of empirical hop count from already known source
Si and the j index of the average score of all closest nodes to the sink. These values are (Hi, j)
Set_Hop_Values (Si, Hi, j) sets the empirical hop count for source Si to be Hi and j to be the index
of the average score of Best_Neighbor set.
Forward (pk, BNk ) forwards the packet pk to the neighbour k which has BNk score
45
01: if (Get_Hop_Values (pk.SourceNode) is Null ) {
02: Forward (pk, BN[1]) // Default forward to best node
03: H ← pk.HopCount
04: Set_ Hop_Values (pk.SourceNode, H, j)
05: }
06: else { //Get_Hop_Values (pk.SourceNode) is not null
07: (H, j) ← Get_Hop_Values (pk.SourceNode)
08: Δh ← H – pk.HopCount
09: index ← j + Δh
10: case ( index ≤ 0 ) {
11: H ← H – index +1
12: index←1 // index of the best node in Neighbor_Set
13: }
14: case ( index > m ) {
15: H ← H – index + m
16: index ←m //index of the worst node in Neighbor_Set
17: }
18: Forward ( pk, BN[index] ) // Smart forward
19: Set_ Hop_Values ( pk.SourceNode, H, j)
20: }
Figure 22: The Smart Greedy Forwarding Algorithm.
Line 7 specifies that we have already an empirical estimation of the hop count H and the
average index j from a particular source. These values are retrieved as shown in line 8. We calculate
(in line 9) the deviation Δh of the hop count of the received packet compared to the stored value H.
The index of the new forwarder neighbour that allows best load balancing will be adjusted by Δh
(line 10). However, two different out of range situations may occur. Line 11 specifies that the
received packet has passed through a lot of hops, and thus it needs to be forwarded to the best node
(i.e. node with index=1). The received packet that has experienced a less hop count than the
empirical value H (line 15), and thus it has to be forwarded to node with higher index (index=m).
The new empirical value is computed (Line 12 and 16) that will be used later as a new reference.
Finally, the packet is forwarded by using the described Smart Greedy Forwarding (line 19).
3.3.2 Walking Back forwarding mode
Because of node failures, node energy depletion due to processing and scheduling activities
and node mobility, disconnections may occur in a WSN generating what we call “voids”. At certain
times, a forwarder node may face a void where there is no closest neighbour to the sink as
illustrated in Figure 25.
46
Score 1
N1
Packet
forwarded
through N1
(the Best)
N2 Nm
Score 2 Score m
Forwarder node
Get_Hop_Values(packet.Source) is Null
Nj
Score j ≈ σ
MeanScore : σ
H ← h
Score [1] > Score [2] > . . . > Score [m]
Recieved packet
h = packet.HopCount
Figure 23: Forwarding the First Packet of a Data Stream.
Score 1
N1 N2 Nm
Score 2 Score m
Forwarder node
(H, j) ← Get_Hop_Values(packet.Source) // not null
Nj
Score j ≈ σ
MeanScore : σ
Score [1] > Score [2] > . . . > Score [m]
Recieved packet
h = packet.HopCount
h = H
h ≤ H+j–m
h =
H+j–2
h ≥
H+j–
1
H ← h–j+1Update: Update:No Update No Update
H ← h–j+m
Set_Hop_Values(packet.source, (H,j))
Figure 24: Forwarding a Packet of an already Known Data Stream.
In this case, the node enters the walking back forwarding mode in order to bypass this void.
In such a case (see Figure 25), the forwarder node will inform all its neighbours that it cannot be
considered as a neighbour to forward packets to the sink. This node will also delegate the
forwarding responsibility to its nearest neighbour to bypass the void. This process does recursively
step back until a node is found that can forward the packet successfully.
47
ForwarderNode Sink
Void
[NO_PATH_TO_SINK] message
Smart Greedy Frowarding[DELEGATE_FORWARDING] message
Figure 25: A Blocking Situation where a Node has no Forwarder Node.
This technique is better than the perimeter routing mode used in GPSR, since this kind of
process is only done once a packet is received from an unknown stream, all the other packets
belonging to the same stream will be routed avoiding the nodes that are facing a void toward the
sink.
3.4 Performances Evaluation
3.4.1 Simulation Environment
We have considered a homogenous WMSN, in which, nodes are randomly deployed through
the sensing field. The sensing field is a rectangular area of 500m x 200m. The sink node is situated at
a fixed point in the righter edge of the sensing field at coordinates (490, 90) while a source node is
placed in the other edge at coordinates (10, 90). We have considered this network for video
surveillance (see Figure 26). In response to an event, the source node will send images with a rate of
1 image per second during 30 seconds. During transmissions, all neighbouring nodes consume hear
the transmitted data packets and thus, nodes consume energy due to this radio communication
behaviour..
Sink
Event Sensor Node
Figure 26: Data Delivery in Response to an Event in a WMSN.
48
To demonstrate and evaluate the performance of our proposed protocol AGEM, we used
OMNeT++ 4 which is a discrete event network simulator [131]. To prove the effectiveness of
AGEM, we have also implemented the GPSR algorithm (as an online but single-path routing
protocol) and an adapted version of MPMPS on top of the TPGF algorithm (as an offline-multipath
routing protocol) and we compared the simulation results. We have also introduced GEAMS
(Greedy Energy-Aware Multipath Stream-based) [132] Routing protocol which consists of a “light”
version of AGEM that does not include the adaptive compass mechanism for next hop node
selection. Thus, GEAMS uses only distance-based greedy progression. Table 1 summarizes the
simulation environment. We have considered that the link data is of type IEEE 802.15.4.
In all simulation scenarios, we assumed that all the nodes are aware of the geographic
location of the sink node. Indeed, such information can be easily embedded into nodes' memory.
However, if thesis not possible other techniques can be used. Authors in [133], have proposed a
distributed discovery mechanism in hybrid wireless networks.
Parameter Value
Network Size
Number of Sink Nodes
Number of Source Nodes
Number of Sensor Nodes
Number of Images
Image Size
Image Rate
Maximum Radio Range
500m x 200m
1
1
30, 50, 80
30 images
10Kb
1 image/sec
80 meters
Table 1: Simulation Parameters (WMSNs).
To evaluate the performance of our protocol, we have considered the following three
topology types:
3.4.1.1 Plain topology:
This topology is used to evaluate the behaviour of the routing algorithm especially the smart
greedy forwarding mode.
Here, we have used three plain topologies; a network of 30, 50 and 80 sensor nodes. An
example of these topologies is shown in Figure 27.
49
Figure 27: A 30-nodes network topology.
3.4.1.2 Topology with holes:
This topology is used to evaluate the performance of the routing algorithm in presence of
holes (i.e. to evaluate the performance of the walking back forwarding mode).
We have used four topologies with holes; a network of 30 sensor nodes with one or two
holes, and a network of 50 sensor nodes with one or two holes. An example of such topologies is
shown in Figure 28.
Figure 28: A 30-nodes network topology with two holes.
3.4.1.3 Regular topology:
This topology is used to evaluate the load-balancing feature of the algorithm. We have used
one grid topology of 26 sensor nodes. This network is shown in Figure 29.
Figure 29: A 26-nodes grid network topology.
In all of the above topologies, we consider the minimum distance between two neighbouring
nodes to be greater than 1 meter. For each topology, we have measured various metrics:
50
— Global Energy Distribution (GED): it is the average and the standard-deviation of
the residual energy at all network nodes.
— Local Energy Distribution (LED): it is the average residual energy in contiguous
regions of 40 meters width.
— End–to–End Delay Distribution: it is the average and the standard-deviation of the
end-to-end delay.
— Packet Loss Ratio: it is the percentage of lost packets during the transmission.
3.4.2 Simulation Results:
In this section, we only present the simulation results obtained for different topologies using
GPSR, TPGF, GEAMS and AGEM. The next section provides the discussion on the results
obtained:
3.4.2.1 Plain topologies
The distribution of the residual energy in the network (GED) is shown in Figure 30.
Figure 30: Average Residual Energy in “Plain” Topologies.
The distribution of the residual energy across the network (LED) is shown in figures Figure
31-Figure 32-Figure 33.
95,5
96,0
96,5
97,0
97,5
98,0
98,5
99,0
30 Nodes 50 Nodes 80 Nodes
Res
idu
al E
ner
gy (
%)
GPSR GEAMS AGEM
95,0
95,5
96,0
96,5
97,0
97,5
98,0
98,5
99,0
30 Nodes 50 Nodes 80 Nodes
Res
idu
al E
ner
gy (
%)
TPGF AGEM
51
Figure 31: Residual Energy Distribution for 30-Node Network Topology
Figure 32: Residual Energy Distribution or 50-Node Network Topology.
Figure 33: Residual Energy Distribution for 80-Nodes Topology.
94
95
96
97
98
99
100
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y (
%)
Sensing Field (m)
GPSRGEAMSAGEM
92
93
94
95
96
97
98
99
100
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y (
%)
Sensing Field (m)
AGEM
TPGF
95
96
97
98
99
100
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y (
%)
Sensing Field (m)
GPSRGEAMS
AGEM
92
94
96
98
100
102
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y (
%)
Sensing Field (m)
AGEM
TPGF
97,5
98
98,5
99
99,5
100
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y (
%)
Sensing Field (m)
GPSRGEAMSAGEM
92
93
94
95
96
97
98
99
100
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y (
%)
Sensing Field (m)
AGEM
TPGF
52
The distribution of the end-to-end delay is shown in Figure 34.
Figure 34: Average end-to-end delay in plain topologies.
The packets loss ratio during image transmission is shown in Figure 35.
Figure 35: Packet-loss ratio in plain topologies.
3.4.2.2 Topologies with holes
The distribution of the residual energy in the network (GED) is shown in Figure 36.
0,00
0,10
0,20
0,30
0,40
30 Nodes 50 Nodes 80 Nodes
Ave
rage
en
d-t
o-d
elay
(t
u)
GPSR GEAMS AGEM
0,00
0,05
0,10
0,15
0,20
30 Nodes 50 Nodes 80 Nodes
Ave
rage
en
d-t
o-d
elay
(t
u)
TPGF AGEM
0,00
2,00
4,00
6,00
8,00
10,00
30 Nodes 50 Nodes 80 Nodes
Pac
ket
Loss
Rat
io (
%) GPSR GEAMS AGEM
53
Figure 36: Average residual energy in topologies with holes.
The distribution of the residual energy across the network (LED) in a topology with holes is
shown in Figure 37.
Figure 37: Residual Energy Distribution across the Network for 50-Node Network Topology with two
holes.
(holes are in region 210m-290m along the sensing field)
The distribution of the E2E delay is shown in Figure 38.
Figure 38: Average End-to-End Delay in Topologies with Holes.
The ratio of overall packet losses during the transmission is shown in Figure 39.
95,50
96,00
96,50
97,00
97,50
98,00
98,50
30 Nodes (1 void) 30 Nodes (2 voids) 50 Nodes (1 void) 50 Nodes (2 voids)
Re
sid
ual
En
erg
y (%
)
GPSR GEAMS AGEM
0.097
0.098
0.098
0.099
0.099
0.100
0.100
50 90 130 170 210 250 290 330 370 410 450 490
Avera
ge R
esid
ual
En
erg
y
(%)
Sensing Field (m)
GPSR
GEAMS
AGEM
0,00
0,10
0,20
0,30
0,40
0,50
30 Nodes (1 void) 30 Nodes (2 voids) 50 Nodes (1 void) 50 Nodes (2 voids)
Ave
rage
en
d-t
o-d
ela
y (t
u)
GPSR GEAMS AGEM
54
Figure 39: The Packet-loss Ratio in Topologies with Holes (Logarithmic Scale)
3.4.2.3 Regular topology
To illustrate the load-balancing feature of AGEM, we have used a grid topology and
simulated a transmission between nodes Src and Dest as shown in Figure 40 (GPSR) and Figure 41
(AGEM). The figures show the residual energy at each node by the mean of a graduated colour that
corresponds to their residual energy (Red to 0% and Blue to 100%).
Figure 40: Residual Energy with GPSR in a Grid Topology.
Figure 41: Residual Energy with AGEM in a Grid Topology.
3.4.3 Simulation Results Discussion
3.4.3.1 Global Energy Distribution (GED)
The GPSR protocol always uses the closest neighbour to the destination (see GPSR
behaviour in a grid topology as shown in Figure 40) due to inflexible selection of the next hop
0,00
10,00
20,00
30,00
40,00
50,00
60,00
30 Nodes (1 void) 30 Nodes (2 voids) 50 Nodes (1 void) 50 Nodes (2 voids)
Pac
ket
Loss
Rat
io (
%)
GPSR GEAMS AGEM
55
node. Forwarding packets to that neighbour is costly since the distance in a greedy forwarding is
considered only and longer the distance is, the most energy consuming the transmission will be.
This explains why residual energy in the case of GPSR is less than in the case of AGEM as shown
in Figure 30 and Figure 36.
Although the use of multiple paths in TPGF, TPGF is still more energy consuming than
AGEM since it uses “greedy” paths.
Moreover, the energy distribution in the network is well distributed with AGEM compared
to GPSR. Unlike GPSR, AGEM use various nodes to perform online multipath routing and load
balancing (see Figure 41).
3.4.3.2 Local Energy Distribution (LED)
Figures Figure 31, Figure 32, Figure 33, and Figure 37 illustrate the average residual energy of
the network partitioned in regions of 40 meters width for the plain topologies and a topology of 50
nodes with two holes. We can clearly see that the energy is uniformly consumed through the
network when using AGEM routing protocol compared to GPSR and TPGF routing protocols.
Moreover, AGEM uses less energy than TPGF since TPGF is a greedy routing protocol and all the
explored paths use always the greedy neighbour to forward packets. The benefit of such a feature is
to prevent the network from being portioned into sub networks that are completely disconnected
if some nodes die because of their energy depletion.
3.4.3.3 Packet Loss and Transmission Delay
By using multiple paths to transmit data packets, not only the packet transmission delay has
been generally reduced first by using GEAMS and AGEM as shown in figures 20 and 26, but also,
this end-to-end delay has become uniform as we can see by the mean of the standard-deviation as
shown in Figure 34 and Figure 38.
However, this end-to-end delay remains quite bigger than the end-to-end delay while using an
offline multipath routing protocol such as TPGF. This can be explained by the fact that TPGF uses
totally disjoint paths to route packets. This makes packets safe from interference problems
(retransmissions).
The packet loss ratio has also been decreased as shown in Figure 35 and Figure 39 in
comparison with GPSR. The decrease in packet loss ratio and delay can be explained by the
following points:
56
— The use of the same path will increase the queuing delays within nodes along the routes
and causes network congestion.
— Sensor nodes have resources constraints, packet loss may occur due to the limited buffer
sizes in sensor nodes.
In the case of topologies with holes, the perimeter routing mode employed by GPSR is not
suited for burst transmissions which causes buffer over loads and packet losses.
These results demonstrate a better performance of AGEM to deliver multimedia traffic (still
images in our simulation case) and provide better QoS compared to GPSR (lower the end-to-end
delay and reduced packet loss ratio). AGEM is also more suitable to dense networks in which
different paths to destination may exist.
3.5 Conclusion
In this chapter, we have described a new algorithm namely AGEM that is suitable for
transmitting multimedia streaming over WMSNs. Because nodes are often densely deployed,
different paths from source nodes to the base station may exist. To meet the multimedia
transmission constraints and to maximize the network lifetime, AGEM exploits the online
multipath capabilities of the WSN to achieve load balancing among nodes.
Unlike classic multipath routing protocols, AGEM routing protocol does not need overall
network topology exploration and paths building before transmitting data packets. With AGEM
forwarding decisions are made online as packets advance towards the destination node. Thus,
control packets are reduced to the minimum.
Simulation results show that AGEM is well suited for WMSNs since it ensures uniform
energy consumption and meets the delay and packet loss constraints.
57
Chapter 4
4 Predictive Routing in Mobile Wireless Networks
4.1 Introduction
Delay/Disruption Tolerant Network (DTN) may often refer to sparse mobile ad hoc
network, where an end-to-end routing path does not necessarily exist. In DTNs, both nodes and
links may be inherently unreliable. Due to these constraints, these networks are referred to as
“challenged networks” [69][74]. Many other emerging communication networks fall into this
paradigm. Vehicular ad hoc networks (VANETs), mobile sensor networks, and nomadic
community networks are few examples.
An interesting DTN example is the city bus network, in which nodes consist of buses (cars,
taxis, trams…) and communicate using short-range radios. With this type of networks, we can
envision a lot of new applications: urban sensing, information dissemination (advertisement, traffic
information, buses software update…) or even Internet access. Since this type of networks does not
rely on an existing infrastructure, and they are formed in an ad-hoc fashion they may be an
excellent solution for information dissemination in certain cases when there is no communication
infrastructure or the existing one is down. Recently, revolution movements have gained some
countries in North Africa. To counteract these protestation movements, government actors have
shut down network infrastructures and disconnected the country from the internet in order to
prevent people from accessing social networks. Many alternatives solutions have risen in the web.
These propositions (such as the OpenMesh Project [134]) all agree to provide mechanisms to
establish networks in ad-hoc fashion and to disseminate easily important information.
The proper functioning of such applications relies essentially on the efficiency of the routing
task. However many challenges affect the routing in DTNs such as the changing network topology
due to intermittent connectivity which is inherent to mobile networks as well as to static networks
(in the case of low duty cycle of the nodes), and it results in low delivery ratio and high end-to-end
delay. The problem of intermittent connectivity can be mitigated if the exact schedule or the
dynamics of the network is known in advance. However, this is not often the case in DTNs as
building this knowledge is an important issue. Thus, the efficiency of a DTN routing protocol
relies essentially on the amount of network knowledge or “oracles” (information about contacts,
queues or even data traffic) available to perform routing decisions.
58
Several routing protocols have been proposed for DTNs. These protocols differ by the
amount of implemented oracles. Depending on the application, some oracles may not be used. For
example, in a city bus network, it may not be possible to embed the entire schedule of contacts
between buses in each node due to various reasons: (1) the memory space needed to store such
information may be of huge size for a communicating node, (2) the “frozen” schedule may not
reflect the actual networks dynamics since the schedule of a bus is not “certain”, (3) the exploitation
of such information (for example, computing the best end-to-end route) can be highly
computational costly. Then, the challenge in designing an efficient routing protocol for such
networks is to make the communicating nodes smarter by using little information about the
network and in a reliable distributed fashion.
In this chapter, we propose ORION [135], a routing protocol for mobile DTNs that
capitalizes on the localization information of the nodes (geo-coordinates) and the nature of contacts
between this type of nodes (buses, cars, taxis, trams) in an urban area. The contribution presented
in this chapter is twofold. First, we have investigated deeply the inter-nodes encounter behaviour.
Second, based on this behaviour analysis, we proposed ORION, a novel routing protocol that relies
on predicting future contacts between nodes and greedy geographic forwarding of data packets.
Thus, with ORION protocol, a communicating node will incrementally build knowledge about its
network regarding the inter-nodes encounters behaviour and nodes positions. Thereby, it should be
able to predict when it will be in contact with other nodes and for how long (duration). In this
chapter, we have also investigated the requirements of ORION protocol in terms of computation
and memory space for time series analysis and forecasting, and storage requirements for bundle
carrying in the context of a store-and-forward routing protocol.
The remainder of this chapter is organized as follows. Section 4.2 presents a brief state of the
art for DTN protocols and the use of stochastic processes and time series analysis in network
communication modelling. Section 4.3 presents details of the proposed ORION protocol. In this
section, we introduce the target application (subsection 4.3.1), and then we present our inter-nodes
encounter behaviour analysis (subsection 4.3.2). Based on these analysis’ results, we define our
routing protocol (subsections 4.3.3 and 4.3.4). Section 4.4 provides extensive simulation results and
related discussion. Finally, section 4.5 concludes the chapter.
4.2 Related Work
In order to overcome the mentioned challenges in DTNs, it is important to design an
efficient routing protocol that uses small network topology knowledge to maximize the delivery
ratio and minimize the delay. Several routing protocols have been proposed for DTNs. These
59
protocols can be classified into two categories; replication-based and prediction (forwarding)-based
protocols. With replication-based protocols, the contacts are assumed to be totally opportunistic
and the required topology knowledge at each node is minimal. In this case, the simplest way to
deliver a message is to send a copy to each encountered node. This is repeated until the destination
receives the message. The Epidemic Routing protocol [66] envisions this strategy. With prediction-
based protocols, only a single copy exists across the network at a given time. The protocol needs to
be supplied with more knowledge about the network. Given the unavailability of topology
information, some protocols try to use probabilities to predict the contact. However, such
prediction can be at the price of reduced delivery ratio. Most of the existing prediction-based
routing protocols focus mainly on whether two nodes would be in contact in the future, without
paying much attention to “when” the contact will happen or “for how long” the contact will last.
This lack of contact timing information degrades the contact prediction accuracy and negatively
impacts the routing performance.
4.2.1 DTN Routing Protocols Taxonomy
As mentioned earlier, the replication-based routing strategy can achieve high delivery ratio
while operating with minimal knowledge. This can be suitable for networks where contacts
between nodes are unpredictable and random. However, this strategy is not optimal in terms of
transmission and buffer size. It also suffers from the lack of scalability. Some protocols, adopting
this strategy cope with this problem by bounding the number of copies in the network trading
delay for buffer occupancy. To limit the replication, two solutions are used:
Fix the number of copies and spread them through distinct nodes. Spray & Wait routing
protocol [80] uses this solution, also called quota-based solution.
Use metrics based on historical encounters between nodes to decide whether to send a copy
or not. PRoPHET [136] (Probabilistic Routing Protocol using History of Encounters and
Transitivity) protocol uses this solution.
The PRoPHET protocol utilizes an algorithm that makes use of the non-random aspect of
the real world. This is done by maintaining a set of delivery success probabilities to known
destinations, and by replicating messages during opportunistic contacts. Replication is done only
for an encountered node which does not have a copy of the message and has a good probability to
deliver the message to its final destination. Given a node i, the probability of node i to encounter
another node j is denoted as P(i,j). The delivery probabilities are computed during each contact
driven by the following three rules:
60
— Updating: ( ) ( ) ( - ( ) )
where is an initializing constant.
— Aging: ( ) ( )
where is an aging constant and is the number of time units elapsed since the last aging.
— Transitivity: ( ) ( ) ( ( ) ) ( ) ( )
where is a scaling constant.
In the prediction (forwarding) based protocols, a node is associated with a forwarding
quality/probability metric for each destination, which is usually a direct (one-hop) forwarding
quality such as contact frequency [136], or time elapsed since last contact [81][85][137].
During a contact, if a node i encounters another node j, node i will decide whether to send
the message to node j based on the comparison between the direct forwarding qualities of node i
and node j. The main drawback of this approach lies in the fact that good forwarding is not
guaranteed due to these observations:
— Node j with a better forwarding quality than node i does not necessary mean that node j is a
good forwarder.
— Despite the good quality of node j, node i may encounter better nodes in the near future.
Similarly, even though the forwarding quality of node j is lower than node i, node j may be
still the best forwarder that node i could encounter in the future.
4.2.2 Times Series in Network Modelling
The proposed ORION protocol makes use of time series to predict contacts. A time series is
an ordered sequence of values of a variable indexed by an ordered set .
The time series analysis serves two purposes: (1) Obtain an understanding of the underlying forces
and structure that have produced the observed data, and (2) Fit a model and proceed to forecasting,
monitoring or even feedback and feedforward control. Time series analysis is used for many
applications such as economic forecasting, sales forecasting, budgetary analysis, stock market
analysis, yield projections, process and quality control, etc. Recently, it starts being used in the field
of computer networks communications. Indeed, time series have gained the attention of many
researchers for the modelling of the Internet and wireless mobile networks traffic. In [138], Basu et.
al. have modelled the Internet traffic using Auto Regressive Moving Average (ARMA) process of
order (p,q). Using this model, they predict the traffic generated by a TCP source using FDDI
protocol. In [139], Liu et. al. have proposed an energy efficient technique for data collection in
Wireless Sensor Networks. A sensor is hold from transmitting redundant data. The data are not
61
sent if they can be predicted by the sink node. For prediction, they utilize Auto Regressive
Integrated Moving Average (ARIMA) model of order (p,d,q) [140] due to its outstanding model fit
and small computational cost. In [141], Herbert et. al. extend this idea to the hierarchic routing
protocol LEACH [142] by providing verification at the cluster head. This approach has shown
great communication cost savings. In [143], Banerjee et. al. used a birth and death process to model
the network’s dynamics. A node entering in the transmission range of a source node is considered
as a birth. Similarly, a death refers to when it leaves this range. Finally, in [144], Singh et. al. extend
this idea by using an AutoRegressive (AR) process to model the number of a node’s neighbours in a
mobile ad hoc network.
When dealing with stochastic processes, values of the involved random variables are taken
over time forming the time series for further analysis. An important step while analysing time
series is to determine the suitable model (or class of models) fitting the observed data. A common
approach to analyse time series is the use of ARMA analysis. An ARMA process is a combination of
an Autoregressive process (AR) and a Moving Average (MA) process. In an AR process, a random
variable is “explained” by its past values rather than other variables. While with MA process, a
random variable is supposed to be explained by its actual mean, augmented by a weighted sum of
the errors (random shocks) that tainted the previous values. ARMA analysis was introduced by Box
and Jenkins [145] and they have identified three steps to model and forecast time series:
— Model Identification: this step is performed to estimate a model structure by using two
essential functions: the autocorrelation function (ACF) and the partial
autocorrelation function (PACF).
— Parameter Estimation: this step is performed for fitting the identified model to the
observed data. This is achieved by determining the coefficients of the linear
combination.
— Forecasting: the final objective is to predict the future values of the time series based
on the already observed data and the linear combination estimated at the second step.
And so, ARMA(p,q) model is defined as: ∑ ∑
∑
∑
(Eq.1)
where:
— Non-negative integers, orders of the AR and MA processes respectively,
— Time-invariant coefficients of the AR and MA models respectively,
62
— Expectation of (often assumed to be equal to zero) and a constant (often
omitted),
— Samples of white noise with mean zero and variance and the white noise
error terms.
To be considered for ARMA analysis, a time series must be stationary. To verify the
stationnarity two conditions must hold:
( ) is constant independent of instant t (Eq.2)
( ) only depends on time lag j. (Eq.3)
4.3 ORION Routing Protocol
4.3.1 Target Application
In this paper, we have considered a city-bus network in which the communicating nodes
consist of buses, trams, cars and hotspots. The buses and trams are assumed to be “regular” mobile
nodes, where the cars are assumed to be “random” mobile nodes and finally the hotspots and access
points to be fixed nodes. The regular nodes move across the area along a certain trajectory, while
random nodes move freely across the urban area. In this scenario, all mobile nodes move with a
non-constant speed. Each communicating node is assumed to be equipped with localization
hardware. Consequently, we propose to utilize geographic addressing and achieve data packets
forwarding in a greedy fashion based on distance and/or angle calculus. With geographic
Routing/Greedy Forwarding, the network address of each node includes geographic coordinates
(for example a node [A], at (x,y) will have the address ID-A.X.Y). Forwarding is then said greedy if
we use the geographic information for choosing forwarder nodes (e.g., choosing the closest
neighbour from the destination node as the next hop).
Following the Box and Jenkins steps to model the times series using ARMA model, we have
conducted some simulations of our city-bus network. In these simulations, instead of using
synthetic mobility models [146][147][148], we studied the two time series in pseudo-realistic
environment mapped on a real city map, namely Bordeaux in France. In this scenario, nodes that
represent bus or trams are moving across the different routes with constant speed between each two
stations. Nodes that represent cars or people are moving freely with constant speed.
ORION is based on greedy geographic forwarding and contact prediction. In the following,
we explain our methodology to achieve an efficient contact prediction.
63
4.3.2 Contact Behaviour Analysis
For our analysis and in order to use efficiently time series, we propose to discrete the time
into small periods of time t . In the description, we further denote ( ) as the
time instants ( ).
In networks with intermittent connectivity, a node becomes aware of an eventual contact by
the mean of periodically exchanged HELLO messages. Consequently, contact (connection) duration
with a certain node is the sum of consecutive periods of time over which the node received at
least one HELLO message from the other node. Respectively, the duration of the non-contact
(disconnection) is the sum of consecutive periods of time over which the node did not receive
any HELLO messages from the other node. The duration of a contact and a non-contact are
two random variables.
In order to study the contact behaviour and based on the consecutive values of the two
random variables and , we construct the two times series and , where denotes
the duration of the ith
contact (connection), and represents the duration of the ith
non-contact
(disconnection). N is the set of natural integers.
Figure 42: Variation of and over time.
Examples of and chronograms are shown in Figure 42. To apply the Box-
and-Jenkins approach, we had to verify the stationnarity of the two time series. Thus, we run the
stationnarity test (see Eq.2 and Eq.3). The results showed that the two stationnarity conditions hold
for almost all the time series obtained from the simulation (at the rate of two times series and
by contacted node at each node). Consequently, and can be analysed using ARMA.
Based on this information, a node can predict the future value of the contact's duration
(connection's duration), and also the future value of the non-contact duration (disconnections'
0
2
4
6
8
10
12
14
16
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Du
rati
on
(ti
me u
nit
)
Connection/Disconnection Index
Disconnection (Non-Contact), Ĉ
Connection (Contact), C
64
duration). Consequently, the node will be able to predict when the next contact will be and for
how long it will last. This knowledge will be extremely beneficial to perform routing decisions.
4.3.3 ORION Contact Model Construction
After running the simulation, we extracted two time series, namely the and ,
for each frequently contacted node. It was interesting to notice that all the time series were quite
similar in terms of pace, even if the mobiles nodes were moving with different non-constant speeds.
The rest of this section describes the Box and Jenkins steps applied to model the proposed
times series. We dubbed this model as “ORION Contact Model” as it is related to the targeted
application scenario.
4.3.3.1 Step 1: ORION Contact Model Identification
We have used Minitab [149] to analyse the obtained time series. The autocorrelation function
(ACF) and the partial autocorrelation function (PACF) are plotted in Figure 43 and Figure 44.
According to the ACF and PACF plots, the results indicate that the best fitting model is the
ARMA(2,1) since PACF presents two significant peaks (i.e. this confirms the AR(2) part), and the
ACF presents one significant peak (i.e. this confirms the MA(1) part).
Figure 43: Autocorrelation Function (ACF) plot for contact's duration.
Figure 44: Partial Autocorrelation Function (PACF) plot for contact's duration.
65
Based on these results, the ORION Contact Model obtained can be written as:
(Eq.4)
where:
denotes the mean value of
denote the ORION Contact Model parameters ( related to the autoregressive
part and related to the moving average part).
are assumed to be independent, identically distributed random variables sampled from
a normal distribution with zero mean ( ) where is the variance.
4.3.3.2 Step 2: ORION Contact Model Parameters Estimation
The second step after identifying the order of the ORION Model is to estimate its
parameters. For the AR part, the parameters can be obtained by the Yule-Walker equations [150].
The principle of the Yule-Walker equation relies on the fact that there is a direct correspondence
between the parameters ( ) and the covariance function of the process. This
correspondence can be inverted to determine the parameters from the ACF which leads to the Yule-
Walker equations:
∑
(Eq.5)
where yielding ( ) equations. is the autocorrelation of Y. is the
standard-deviation of the input noise process, and the is the Kronecker Delta function. This
equation is usually solved by representing it as a matrix, getting the following equation solving all
.
3
2
1
012
101
210
3
2
1
(Eq.6)
This equation provides a way to estimate the AR(p) parameters by replacing the theoretical
covariance with estimated values. For the MA part, the single parameter is obtained by
identification based on the estimated AR parameters and the last estimation error.
4.3.3.3 Step 3: Forecasting
Since the orders of the ORION Model are fixed in time due to the mobility model, the
computational cost of resolving linear systems can be avoided by extracting generic formulas. The
66
node will have to compute the model parameters based on simplified mathematical expressions.
Moreover, since these formulas include only aggregated data (sums, means, standard-deviations,
variances …), there is no need to store all the past data; only few values are maintained at each node.
The following sub-section explains the different steps to derive these simplified mathematical
expressions and how they will be used for forecasting. Since the model is ARMA(2,1), it is
composed of two parts: AR(2) and MA(1).
For AR(2) process we have :
We multiply both sides by one lag
value and take the expectation:
We eliminate the zero correlation
forcing term (<xtξt+1>) and we
divide by N – 1:
We then have: (ci is the covariance or lag i)
We divide both sides by c0, we
obtain the equation : since
Doing the same thing with lag2, we
find the second equation. Then we
have :
Leading to the Yule-Walker
equations:
, and so
Knowing that r0=1 :
Finally, we get :
an
d
Considering the MA(1) process:
(where ξ is the white noise N(0,σ²))
We multiply both sides by one lag
value and take the expectation:
We eliminate the zero terms and
we divide by N – 1 :
(knowing that
)
And we get :
By repeating the same operations
for zero lag value :
We obtain :
1 1 2 1 1t t t tx x x
1 1 2 1 1t t t t t t t tx x x x x x x
1 11 21 1 1
t t t t t tx x x x x xN N N
1 1 0 2 1c c c
1 1 0 2 1r r r 0i ir c c
1 1 0 2 1
2 1 1 2 0
r r rr r r
0 11 1
1 02 2
r R
r rrr rr
2 2 2 2 2 22 2 2 22 2 1 0 1 0 1 0 1 02 2 2 2 2 2 2 21 02 21 0 1 02 21 0 1 02 21 0 1 02 21 0r r r r r r r r1 0r r1 0 1 0r r1 0 1 0r r1 0 1 0r r1 02 2r r2 2 2 2r r2 2 2 2r r2 2 2 2r r2 21 02 21 0r r1 02 21 0 1 02 21 0r r1 02 21 0 1 02 21 0r r1 02 21 0 1 02 21 0r r1 02 21 01R r
1 1 12
2 1 21
1111r r
r rr
1 21 2
1
11
r rr
22 1
2 211
r rr
1t t tx
1 2 1 1t t t t t tx x
2 21 2 1 2 1 1t t t t t t t t tx x
211
1 1tt tx x
N N
2
0 if if 1
i j i ji jN
21c
1 1t t t t t tx x
2 2 2 2 20 0 1c c
67
Knowing that, and from the two
obtained equations we have :
By solving the quadratic equation
we find :
From the three boxed formulas, we can see that the estimation of ORION Contact Model
parameters relies on the estimated values of autocorrelation of order 1 and 2. (i.e. r1 and r2).
Statistically, autocorrelation of order p (i.e. rp) is estimated by the following expression:
where:
;
;
;
We can see that the calculus of the autocorrelation terms relies on aggregated terms
(averages). Thus, these values can be incrementally computed without keeping the entire data set as
follows:
Finally, in order to adaptively estimate the parameters of the ARMA(2,1) process and in real-
time, we only need to store the following 7 variables (rather than the entire time series):
— The number of the considered values so far [0, t)
— The last three values of the time series.
— The last average of the values (at t-1)
— The last average of the squared values (at t-1)
11 2
0 1
cr
c
2
1
1
11 2
2r
r
22
t t p t t p
p
t
E X X E X Xr
Var X E X E X
1
1
1
1
N
X t i
i
E E X xN
2
2 2
1
1
1
1
N
t iXi
E E X xN
,
1
1
1p
N
X X t t p t t p
i
E E X X x xN
( 1)
( )
( 1)t
t
X t
X
N E xE
N
2( 1)
2( )
2( 1)t
t
tX
X
N E xE
N
( 1)
( )
,
,
( 1)p t
p t
X X t t p
X X
N E x xE
N
N
1 2, ,t t tx x x
XE
2XE
68
— , The last values of and respectively
Since the data are evolving in time, we propose to compute these parameters in an
incremental fashion. For two different instants t1 < t2 the estimated parameters are different
because the estimation at t1 takes into account the data up to t1 (i.e. [0, t1]) and similarly the
estimation at t2 takes into account all the data in [0, t2]. This approach makes the estimation in real-
time and more accurate with new observed data. The process of model parameters’ update and
future value forecasting is explained with the pseudo code in Figure 45.
On_New_Value(v)
010: N++;
020: Update_Sliding _Window (v, xt, xt-1, xt-2);
030: EX = Compute_New_Average (EX, N, xt);
040: EX² = Compute_New_Average (EX², N, xt);
050: EX,X1 = Compute_New_Average (EX,X1, N, xt, xt-1);
060: EX,X2 = Compute_New_Average (EX,X2, N, xt, xt-2);
070: r1 = Compute_AutoCorrelation (EX, EX², EX,X1);
080: r2 = Compute_AutoCorrelation (EX, EX², EX,X2);
090: φ1 = Compute_Phi1 (r1, r2);
100: φ 1 = Compute_Phi2 (r1, r2);
110: θ = Compute_Theta (r1);
120: xt+1 = Estimate_Future_Value__ARMA__2_1 (φ 1, φ 2, θ);
Figure 45: Online model parameters update, and future value forecasting
Figure 46 presents a comparison between forecasting based on “offline” parameters
estimation (i.e. the parameters are estimated considering the entire data set, then the future values
are computed at each index using the obtained model) and forecasting based on “online” estimation
(i.e. the parameters are estimated for each new observation, considering the available data so far).
We can clearly see that our online estimation is better than the offline estimation in predicting the
actual data.
1,X XE2,X XE
1,X XE2,X XE
69
Figure 46: Forecasting with "online" and "offline" parameter estimation.
4.3.4 Forwarding Algorithm
The forwarding algorithm used in ORION is based on three criteria: (1) in order to forward
a packet in a greedy manner, a node will look for the closest connected neighbour to the
destination, (2) if such a node is not available, the forwarder node will look for the most advancing
connected neighbour toward the destination, and finally (3) if there is no such a node, the
forwarder node will schedule the data packet for the best future connected neighbour. A pseudo
code of the algorithm is shown in Figure 47. A detailed version is shown in Figure 48.
On_Forwarding_Packet
10: nextHop ← Closest Connected Neighbour to the Destination;
20: IF (nextHop ≠ null) GOTO 80;
30: nextHop ← Most Advancing Connected Neighbour towards the Destination;
40: IF (nextHop ≠ null) GOTO 80;
50: nextHop ← Best Future Connected Neighbour;
60: IF (nextHop ≠ null) GOTO 90;
70: Store_Packet(); END.
80: Send_Packet_To(nextHop); END.
90: Schedule_Packet_For(nextHop); END.
Figure 47: Pseudo code for ORION forwarding algorithm.
-5
0
5
10
15
20
25
30
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76
Co
nta
ct
Du
rati
on
(T
ime
Un
its
)
Contact Index
Actual DataForecasts with "online" estimationForecasts with "offline" estimation
70
Forward_Packets (pk)
Packet_Queue: a data structure where all the packets to be sent is stored.
Connected_Neighbors_Set (CN): the set of all neighbours that are currently in contact with this node.
Estimated_Neighbors_Set (EN): the set of all estimated neighbours, i.e. the nodes whose contacts we
have a historical data about.
Forwarder_Node (FN): the address of the next hop.
: This function gives a score to a neighbour based on its next contact date. This function can be
configured to give priority to delivery speed or delivery certainty, or both of the two criteria.
01: if (Packet_Queue is not Empty) {
02: pk= Packet_Queue.pop();
03: if (pk.Next_Hop == null) { //packet was not scheduled
04: ‖ ‖
05: if (FN ≠ null)
06: send_packet(pk,FN);
07: else { // Most advancing neighbour towards the destination
08: (‖ ‖ ‖
‖)
09: if (FN ≠ null)
10: send_packet(pk,FN);
11: else { //best future contact
12: ( )
13: pk.Next_Hop = FN;
14: Packet_Queue.push(pk);
15: }
16: }
17: }
18: else { //packet was scheduled for a certain node
19: if (pk.Next_Hop ){ // is the predicted neighbour connected?!
20: ‖ ‖
21: send_packet(pk, FN);
22: }
23: else { // the predicted neighbour is not connected
24: pk.Next_Hop = null;// unscheduling the packet
25: Packet_Queue.push_back(pk); // storing the packet
26: Forward_Packets(); // Starting Over
27: }
28: }
29: }
Figure 48: A detailed description of the ORION forwarding algorithm.
71
4.4 Performances Evaluation
4.4.1 Simulation Environment
For simulation purpose, we have considered a homogenous wireless mobile network in
which nodes are randomly deployed through an area of 1000m x 1000m. Two nodes are selected
randomly at the beginning of the simulation to act as source and destination. The source sends
periodically data packets to the destination. The simulation is run for 1800 seconds (letting
sufficient start-up time for PRoPHET, i.e. 600 seconds). To demonstrate and evaluate the
performance of ORION, we used OMNeT++ 4.0 [131]. As a comparison term, we use the
PRoPHET protocol. We considered variant network topologies by varying (1) the number of
nodes (i.e., 30, 50, and 70 nodes) and (2) the nodes speed (i.e., 5 m/s, 10 m/s, 15 m/s and 20 m/s).
For each topology, we measured various parameters: (1) the average hop count from the source to
the destination, (2) the packets delivery ratio, (3) the first packet arrival, and finally (4) the average
end-to-end delay. To avoid redundancy, results relative to the 50 nodes topology are not shown
since they are similar to those of the 70 nodes topology.
We also prove that ORION routing protocol does not require more space memory to
perform store-carry-and-forward routing by measuring the message queue occupancy during all the
simulation period.
4.4.2 Simulation Results Discussion
4.4.2.1 Hop Count (HC)
Figure 49: Average hop count in 30 and 70 nodes topologies with variant speed.
From Figure 49, we can clearly see that ORION delivers packets along fewer hops than
PRoPHET and this is the case for all the three topologies and with all nodes speeds. This is
achieved thanks to the twofold forwarding strategies of ORION protocol (store-and-forward and
store-carry-and-forward) while PRoPHET is just a store-and-forward protocol.
0
5
10
15
20
5 m/s 10 m/s 15 m/s 20 m/s
Ho
ps
Speed
30 Nodes - ORION
30 Nodes - PRoPHET
70 Nodes - ORION
70 Nodes - PRoPHET
72
4.4.2.2 Packet Success Ratio (PSR):
Figure 50: Packet Success Ratio in 30 and 70 nodes topologies with variant speed.
Since the selection of the next forwarder node in ORION is based on three criteria (i.e.
closest neighbour, most advancing neighbour, and the first future contact) rather than just one
criterion (Probability of delivery success) in the case of PRoPHET, the successful node selection in
ORION prevents packets from being lost; i.e., sent to nodes that cannot forward them. This allows
ORION to successfully deliver more packets than PRoPHET as shown in Figure 50. From this
figure, we can also notice that the impact of nodes' speed is more important in ORION than in
PRoPHET. With a high speed, the accuracy of the ARMA predictions is affected since the contacts’
durations will be at the same scale as prediction error margin. Thus, packets loss will be more
frequent. However, the packet delivery ratio is still higher than Prophet’s.
4.4.2.3 First Packet Arrival (FPA) and End-to-End Transmission Delay (EED):
Figure 51: First Packet Arrival in 30 and 70 nodes topologies with variant speed.
0
15
30
45
60
5 m/s 10 m/s 15 m/s 20 m/s
Rati
o (
%)
Speed
30 Nodes - ORION
30 Nodes - PRoPHET
70 Nodes - ORION
70 Nodes - PRoPHET
0
30
60
90
120
5 m/s 10 m/s 15 m/s 20 m/s
Tim
e (
tu)
Speed
30 Nodes - ORION
30 Nodes - PRoPHET
70 Nodes - ORION
70 Nodes - PRoPHET
73
Figure 52: Average E2E Delay in 30 and 70 nodes topologies with variant speed.
Because of the greedy nature of ORION, packets will always choose either the shortest or
the "earliest" next hop making packets arrive more quickly at the destination node (Figure 51) and
experiencing shorter end-to-end delay (Figure 52) compared to PRoPHET where the next hop is
chosen based on only its success delivery probability.
4.4.2.4 Bundle Queue Occupancy:
Figure 53: Maximum message queue occupancy in 30 nodes topologies with variant speed.
From Figure 53 we can clearly see that ORION routing protocol does not require excessive
memory space for message queue in order to achieve store-and-forward routing. In our simulations,
and in sparse network topology (30 nodes), the maximum message queue occupancy is high (~26
messages in case of 5 m/s node mobility speed) when the node mobility speed is slow. Due to the
rarity of the contacts, nodes are obliged to keep messages for a long period of time. However, with
higher node mobility speeds, the maximum queue occupancy decreases accordingly (maximum of
~5 messages in the case of 20 m/s node mobility speed). Furthermore, in dense network topology
(70 nodes), the maximum message queue size is bound to ~2 messages which it is more than
acceptable.
0
5
10
15
20
5 m/s 10 m/s 15 m/s 20 m/s
Tim
e (
tu)
Speed
30 Nodes - ORION
30 Nodes - PRoPHET
70 Nodes - ORION
70 Nodes - PRoPHET
0
5
10
15
20
25
30
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Bu
nd
le C
ou
nt
Host
5 m/s 10 m/s
15 m/s 20 m/s
74
4.5 Conclusion
In this chapter, we have described a new routing protocol, dubbed as ORION, which is
suitable for mobile delay tolerant networks. Since the network dynamics are often not so random
such as in city-wide inter-hotspot network interconnected through taxis, buses and vehicles, the
contacts between two communicating nodes can be analysed and, moreover, predicted. ORION
routing protocol is based on greedy geographic forwarding and contacts predictions, switching
between store-and-forward and store-carry-and-forward strategies in such a way, that packet
forwarding is always optimal. ORION uses ARMA model online parameter estimation to predict
future contacts due to its outstanding fit to this kind of network dynamics. Simulation results show
that ORION routing protocol outperforms PRoPHET in terms of different metrics such as first
packet arrival delay, end-to-end transmission delay, hop count, and Packet Delivery Ratio.
Moreover, ORION routing protocol does not require excessive memory space to achieve efficient
delay-tolerant routing.
75
Chapter 5
5 Stochastic Topology Control in Wireless Sensor Networks
5.1 Introduction
The success of some applications of wireless sensor networks depends greatly on the aimed
network topology, since it will impact the main tasks such as the routing protocol and the data
collection scheme, but also the energy consumed for radio communications.
Considering various applications of WSN (for example, battlefield surveillance), sensor nodes
can rarely be deployed in a deterministic way. Thus, random node deployment is often the only
solution. However, such deployment has some disadvantages. First, if dropped, sensor nodes are
not guaranteed to be still functional once on the ground. Moreover, some embedded hardware like
those for localization (Positioning systems, Angle measurement, etc.) or even the hardware for
communication (i.e. the radio antenna) can be damaged or will not function properly. Second, the
network topology is unpredictable and this will impact heavily all the networking tasks such as
node organisation, data gathering, packet forwarding, etc. Consequently, and in order to ensure the
desired application, nodes may consume more energy in various tasks:
(1) Nodes may have to transmit the data packets over long distances and thus consume more
energy in communications, as shown in Figure 54.
(2) A node may become a cluster-head in a hierarchical organization, with less capability, less
residual energy, etc.
(3) A node may become a hot spot in order to keep the global connected topology from being
partitioned, as shown in Figure 55.
Figure 54: Transmission over Long Distances Figure 55: Hot Spot Node
76
These network topology issues have received a lot of attention in the literature. On the one
hand, a lot of research works have been carried out in order to optimize the WSN applications by
making them aware of the underlying network topology. Thus, a lot of routing protocols have
taken advantage of geographic information (coordinates, angle, distances, etc.) in order to make
packet forwarding decisions. These geographic routing protocols rely on greedy forwarding to
route the data packets from the source to the final destination [49][121] with high efficiency and
low energy consumption. A lot of these protocols addressed the problem of topology holes that
may be formed due to various reasons (nodes' deployment, nodes' mobility, nodes' failure, etc.).
Techniques such as face routing in planar graphs [53] or step-back-and-mark [151] have been used.
On the other hand, the trend is for approaches that optimize the WSN using topology control.
These approaches act on the topology. Towards this objective, we can find techniques where nodes
adjust their transmission range in order to adjust the one-hop neighbourhood size
[44][152][153][154][155]. In some other techniques, node may act on their wake/sleep scheduling
[50][54][156][157] to achieve an optimized network topology where lesser nodes are active, and thus
maximizing the global network lifetime. These topology adjustment approaches can be divided into
two main categories: (a) sensing coverage topology control, and (b) connectivity topology control.
The first category aims to optimize the surface area being covered by the node's sensing hardware
(for example, the angle of view of a camera). The optimization concerns the maximization of the
surface being sensed, but also the minimization of the number of sensor being involved
[158][159][160][161][162][163][164]. While, the second category aims to optimize the
communication infrastructure by maintaining the nodes connected and keeping the whole network
from being partitioned into disconnected clusters [44][50][54][152][153][154][155][156][157][165].
5.2 Related Work
Topology issues have been widely studied in the literature, a coherent taxonomy is given in
[166]. Topologies control problems are divided into two main categories: (1) Sensor coverage
topology, and (2) Sensor connectivity topology. The first one is concerned about maximizing the
sensing area while consuming less energy, whereas, the second one is concerned about network
connectivity.
5.2.1 Sensor Coverage Topology
Sensor coverage can be studied in static, mobile and hybrid networks.
— Static Networks: all the nodes are static and their location is a direct result of the node
deployment. Many approaches have been proposed targeting different levels of
sensing coverage: (1) partial sensing coverage [164][167], (2) single sensing coverage
77
[168], and (3) multiple sensing coverage [160][162][163]. In partial coverage, only
partial area is sensed within a certain period of time. In single coverage, nodes try to
sense non overlapping areas in order to cover the entire sensing field. Finally, in
multiple coverage, an area is covered redundantly by more than one node.
— Mobile Networks: nodes are mobiles and move to their best location in order to
optimize the sensing coverage. Works in [158][159][161] studied the best relocation
schemes in order to achieve efficient sensing coverage.
— Hybrid Networks: only some nodes are capable of moving. They can help the
achievement of a desired coverage by moving to their best computed locations.
Authors in [168] proposed a combined solution for the exploration and the coverage
of a given area. Authors in [170] considered single sensing coverage problem by
relocating mobile nodes to overcome topology holes.
5.2.2 Sensor Connectivity Topology
In this context, energy can be reduced according to two main schemes. The first one
concerns the adjustment of the transmit power, resulting in adjusted transmission range. The
second scheme concerns the decision making about which nodes should be turned on/off, when,
and for how long.
— Power Control Mechanisms: the goal is to dynamically change the transmitting range
in order to maintain some property of the communication graph, while minimizing
the node's energy consumption. Works in [44][152][153][154][155][165][171] fall into
this category.
— Power Management Mechanisms: the goal is to decide which of the nodes should be
turned on/off and when in order to build an energy saving topology and thus to
maximize the network lifetime. Works in [50][54][156][157] fall into this category.
5.2.3 Discussion
Wireless sensor networks are a very promising technology. Indeed, the philosophy behind
this technology is that with lower costs, WSN-based applications can perform as well as
applications that rely on solid in
A lot of WSN applications do not require the use of mobile sensor nodes although it may
enhance greatly the considered application. Making the sensor nodes mobile capable require the
embedding of a whole different hardware (wheels, engines, etc.) and require more energy (mobility
hardware is much energy consuming than any of the classic nodes' functions). Moreover, such
78
sensor nodes will cost mush more than static sensor nodes which will be against the WSN
philosophy. However, some applications (critical, harsh environments, etc.) cannot do without it.
Sensing range is often distinct from the radio communications range. Thus building a
network topology for optimizing the coverage differs depending on the coverage type to maximize
(sensing or communication). Regarding the targeted application, certain coverage can be more
important than the other. For applications where nodes do transmit data very often, it is required
to optimize the packet forwarding infrastructure, since radio communication is the most energy
consuming task in wireless sensor networks.
In some WSN-based applications, sensor nodes are randomly and densely deployed which
may be very inefficient in terms of data gathering, network lifetime, etc. Indeed, dense deployment
present some issues such as radio jamming that may make the nodes retransmit their data packets
very often if the MAC layer handles retransmissions, else communications will suffer from severe
packet loss. Another issue concerns the network lifetime, if no energy management mechanism is
implemented, nodes will be always on a wake state. This lead to the fact that a large number of
nodes will be involved for a packet forwarding that may concern just few nodes, since all the
neighbouring nodes are hearing the radio transmissions. Moreover, such deployment will increase
the delays due to the scheduling effort required for the transmissions' synchronization in a dense
deployment.
Wireless sensor networks are often densely deployed, and an efficient packet forwarding
infrastructure is thus required in order to maximize the network lifetime by making the nodes live
longer. Since the radio communication task is the most energy consuming, such infrastructure can
be achieved through an efficient on/off schedule of the radio hardware at each node. The challenge
is then to design a distributed nodes' activity schedule that tries to put a large number of nodes in
an off state while ensuring that every data packet can still be successfully forwarded to its
destination, i.e. the obtained topology still offers a path from the source to the final destination of
the packet.
5.3 System Model
In order to design an energy-efficient activity schedule, we need to address the following
points:
(1) What is the algorithm to run at each sensor node in order to decide whether to put the
radio hardware in off state (entering the sleep mode)?
(2) How long should a sensor node remain in the sleep mode?
79
5.3.1 Node Deployment
In the following, we consider a predetermined sensing field (we consider rectangular sensing
field for simplification, having a length of L and a width of W). N nodes are deployed randomly
across this sensing area. All sensor nodes are homogeneous in terms of sensing hardware,
communication range and initial energy provision. The radio transmission range of each node is
equal to R. Radio transmission range is uniformly a disk with a radius equal to R. Thus, the
generated communication graph is a unit disc graph.
We call "neighbour" of node i, every node j such as the distance between i and j is lower than
R. the number of one node's neighbours depends on the initial node deployment mode and the
sensing field's surface.
Figure 56: Neighbouring Probability in Uniform Deployment
For uniform deployment, the average number of neighbours can be computed as follows:
The probability for a node j to be a neighbour to node i can be expressed as the probability
of the node j being deployed inside the transmission range of node i, i.e. the coordinates of node j
(xj, yj) fall within the disc having as centre the node i coordinates (xi, yi) and R as its radius (see
Figure 56). Thus, this probability can be:
( )
(5.1)
( )
(5.2)
Thus, the average number of neighbours, , is equal to:
( )
(5.3)
80
However, experimental data shows results that differ from theoretical results due to the
border effects [172]. Indeed, nodes that are located at the border of the rectangular sensing field
have fewer neighbours than nodes that are fully located within the sensing field. The average
neighbours count has been investigated in [173] through simulations. Authors conclude that, if
border effect is negligible, a binomial approximation is enough accurate to estimate the
neighbourhood size. Works in [172][174] show that the average neighbours count can be
approximated by the following formula:.
( )
(
) (5.4)
5.3.2 Wakeup/Sleep Schedule
For random node deployment, the average number of neighbours is considered uniform and
it equals . In order to decide the duration of each state: sleep and awake, a node can simply choose
durations according to two any statistical distribution such as Poisson, Pareto, Log-Normal, etc. the
benefit from such duration selection method is that each node can choose its values without any
additional knowledge about its neighbourhood. In the following, we analyse the consequences of
such schedule in terms of neighbourhood size.
We consider a WSN where all nodes are deployed uniformly across the sensing field. Every
node is choosing its wake-up duration according to a Poisson distribution with parameter ,
followed by a sleep period of a duration that follows a Poisson distribution with parameter .
Figure 57 represent the number active neighbours in this context.
Figure 57: Number of Active Neighbours
81
The wakeup/sleep process is an On/Off Markov Process , having the following
transitions probabilities: ( ) ( ) , and ( ) ( ) (a
wakeup state is automatically followed by a sleep state and vice versa). The average wakeup state
duration is ( ) , and ( ) . This leads to the following infinitesimal generator:
(
) (5.5)
Solving the equation let us find the stationary probability that a node is on a sleep
state or a wakeup state:
and
(5.6)
If we consider the superposition of M homogenous and independent On/Off Markov
processes { }, we will have the aggregated process ∑ taking its
states in . State 0 is when all the processes are on OFF state and M when all the processes
are on ON state.
— If the process is in the state , the only possible transition is corresponding to
the end of one of the OFF periods, so the average duration in this state is ( ) .
— Respectively, if the process is in the state , the only possible transition is
corresponding to the end of one of the ON periods. So the average duration in this state is
( ) .
— In the state , the next possible transitions are:
, if there is an OFF period ending among the OFF periods at this
instant.
, if there is an ON period ending among the ON periods at this instant.
Thus, the duration of the state is the minimum of two variables A and B, the first one
of parameter ( ) that corresponds to the transition towards and the second one of
parameter that corresponds to the transition towards . Thus, ( ) ( ) .
Then, we have the following transitions probabilities:
— ( )
— ( )
— For :
( ) ( ) ( )
( )
82
( ) ( )
( )
Leading to the following infinitesimal generator:
(
(( ) )
( )
(( )( ))
( )
)
(5.7)
To get the stationary probability distribution , we solve the equation . Putting
, we get:
( ) (5.8)
( ) (5.9)
5.3.3 Target Wakeup/Sleep Schedule
Thus, the probability to have an average number of k active neighbours among an average of
M neighbours can be computed following (5.4). In a densely deployed network, it would be
profitable if the maximum number of nodes can be put in a sleep state without disconnecting the
nodes. Towards, this objective, reducing the size of active neighbourhood seems promising. Indeed,
to forward data packets, a node need only one neighbour. However, since every neighbour is not
always located in the best direction towards any destination node, reducing the active
neighbourhood to a smaller set is more appropriate.
The question that arises after we have computed the probability to have i active neighbours
among a total of M neighbours, is how to maximize this probability. This probability function is
represented in Figure 58.
83
Figure 58: Probability to have K neighbours
The probability function ( ) exhibits a maximum at the point:
(5.10)
Thus, the best wakeup/sleep ratio ( ) in order to maximize the probability to have an
average of k active neighbours among an average of M neighbours is the value .
In order to optimise the active neighbourhood size, a general approach is to tune the
wakeup/sleep schedule in such a way to maximize the probability to have the active
neighbourhood size contained in a certain interval [ ], with (in order to not disconnect the
node from the network) and (in order to use less neighbours than the actual neighbours
count). This probability, [ ], can be expressed as the following:
[ ] ( ( ) ) ∑
(5.11)
[ ] [ ] ∑
( )
(5.12)
[ ] [ ]
( ) ∑
(5.13)
84
The probability function [ ]( ) is shown in Figure 59. This function exhibits a
maximum for the value [ ]
. We can see that this function includes a partial sum of the binomial
formula ( ) . It is worth to notice that no closed form exists for such sum.
Figure 59: Probability to Have [u,v] Active Neighbours
Finding [ ]
requires solving for the equation:
[ ]( ) (5.14)
( ) ∑
( ( )) (5.15)
∑
( ( )) (5.16)
Solving this equation is equivalent to find the root of a ( )–degree polynomial. This
equation may be complex to solve when the degree of the polynomial increases. Thus,
approximation techniques can be. By using the Newton's method, the solution to this equation can
be found very quickly.
Given the function and its derivative :
85
( ) ∑
( ( )) (5.17)
( ) ∑
( ( )) (5.18)
Then, an efficient process for calculating a good approximation of the value [ ]
, is:
( )
( ) (5.19)
5.3.4 Local Topology Awareness and Path Construction
We consider WSN-application where nodes are organized into a flat topology. Every node
implements a heart-bit application designed to advertise its distance towards the sink node in terms
of number of wireless hops.
Every node keeps in its local memory information about how to reach the sink node. This
information include its distance towards the sink node in terms of hop count (HopsToSink) and the
neighbour (NextHop) through which to send all data packets addressed to the sink node (packets
generated locally or packets to be relayed).
BeaconPacket = {SenderID, HopsToSink}
At the start-up of the application, the sink node broadcast a beacon packet containing a field
"HopsToSink" equal to 0. At the reception of this packet, every node will update its information on
how to reach the sink node if a best forwarder node is found. If not, the node's local information
are not updated. The node will then broadcast its beacon packet with the up-to-date information.
Figure 60 represents this process.
Upon_Receiving_a_Beacon( pk )
Parameters:
HopsToSink: a variable to store the distance to the sink node in terms of hops
(initially set to -∞)
NextHop: the forwarder neighbour through which to reach the sink node
(initially set to Ø).
86
if (HopsToSink = -∞) then
goto UPDATE; // update on the reception of the first beacon
if ( (pk.HopsToSink+1) < HopsToSink) then
goto UPDATE; // update on a better forwarder
goto EXIT;
UPDATE: HopsToSink = pk.HopsToSink + 1;
NextHop = pk.SenderID;
EXIT: end;
Figure 60: Path Construction Algorithm
Figure 61 shows an example of path-to-sink building. In this network, every node knows its
next neighbour to use in order to forward packets to the sink node.
Figure 61: Example of Path Construction
The beaconing process can be executed periodically in order to deal with topology changes
that happen due to nodes' state switching (sleep/wakeup states).
In general, the process of sending periodical beacon messages allows wireless nodes to be
aware of their local neighbourhood. However, this awareness depends on the message exchange
frequency. For static networks with sleep/wakeup state scheduling, the exchange of such messages
can be reduced to a minimum and can be activated only when a network topology change occurs in
the local neighbourhood. The beaconing process may be triggered at a wider neighbourhood than
the initial one-hop neighbourhood if necessary. Finally, paths that may have been broken when a
node switches to sleep state can thus be recovered rapidly. Figure 62 shows an example of the path
repair mechanism results. The green arrows are the established paths to the sink using the selected
NextHop at each node. When an intermediate node switches to the sleep state, the local
neighbourhood becomes aware of this, and starts the repair mechanism by sending again beacon
messages advertising their distance towards the sink node. Nodes that were using the node "S" do
87
not have a forwarder now, so their local information (HopsToSink, NextHop) will be set to their
initial values (-∞, Ø). The beaconing process is activated progressively wider, until all nodes will
have a path to reach the sink.
Figure 62: Example of Path Recovery
5.3.5 Data Packet Forwarding
Since paths are set up in a more static way, data packets forwarding is made simple. Indeed,
every time a node receives a data packet that needs to relay, this node will forward the packet
through its NextHop neighbour. Whenever, this neighbour is not available (neighbour node
entering sleep state, or neighbour node failure) the data packet is then kept into a queue
(PacketQueue) waiting for the path repair mechanism in order to select a new neighbour for
forwarding data packets to the sink node. Figure 63 represents this forwarding algorithm.
Upon_Receiving_a_DataPacket( pk )
Parameters:
HopsToSink: a variable to store the distance to the sink node in terms of hops.
NextHop: the forwarder neighbour through which to reach the sink node.
PacketQueue: a queue to store all outgoing data packets, ready to be sent.
ID: the identifier or the local wireless node.
if (pk.DestinationID = ID) then
ProcessPacket(pk); // packet reached its destination
goto EXIT;
if (NextHop≠ Ø) then
SendPacket(pk, NextHop); // packet forwarded
goto EXIT;
LATER: PacketQueue.Push(pk); // sending delayed until path recovery
EXIT: end;
Figure 63: Packet Forwarding Algorithm
88
5.4 Performances Evaluation
5.4.1 Simulation Environment
In order to prove the performance of our stochastic wakeup/sleep scheduling, we have
considered a homogenous MWN, in which, nodes are randomly deployed through the sensing
field. The sensing field is a rectangular area of 1000m x 500m. The sink node is situated at a fixed
point in the righter edge of the sensing field at coordinates (980, 250). We have considered this
network for a monitoring application with an event-driven data gathering scheme. In response to
an event, the involved node will send a data packet to inform the sink node about the occurred
event.
To demonstrate and evaluate the performance of our proposed mechanism, we used
OMNeT++ 4 which is a discrete event network simulator [131]. Table 2 summarizes the
simulation environment.
Parameter Value
Sensing field
Transmission Range
Number of sink nodes
Number of nodes
Wakeup/Sleep ratio (ρ):
Wakeup Rate (λ)
Sleep Rate (μ)
Average Data Event frequency (δ)
1000m x 500m
R=120m
1
N={100, 150, 200, 250}
ρ={ρ2, ρ3, ρ4, ρ20%, ρ40%, ρ60%, ρ80%}
λ={ λ2, λ3, λ4, λ20%, λ40%, λ60%, λ80%} s-1
μ=0,01 s-1
δ=200s
Table 2: Simulation Parameters (Topology Control)
The following table presents the simulation parameters related to the wakeup/sleep
scheduling. For each node count (N), an average neighbour count (M) is calculated according to the
formula described in (5.4) due to random deployment. For each configuration (N, M), we
considered different values for the ratio ρ. These values have been chosen in such a way to have 2,
3, 4 average awake neighbours and generally 20%
, 40%
, 60%
, 80%
of the total neighbours. These values
are consolidated in Table 3. For example, in the defined sensing field, and using random
deployment of 200 nodes, the best value for the wakeup/sleep ratio (ρ) to have the best probability
to in order to have an average of 3 awake neighbours is ρ=0.20, i.e., a sleep ratio that is five times
greater than the wakeup ratio.
89
N ~M1 ~M2 ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
100 9 8 0,29 0,50 0,81 ρ2=0,21 ρ3=0,52 ρ5=1,05 ρ6=2,14
150 13 11 0,17 0,29 0,42 ρ2=0,21 ρ5=0,52 ρ7=1,05 ρ9=2,14
200 18 15 0,12 0,20 0,29 ρ3=0,21 ρ6=0,52 ρ9=1,05 ρ12=2,14
250 23 19 0,10 0,15 0,22 ρ4=0,21 ρ8=0,52 ρ12=1,05 ρ15=2,14
Table 3: Simulated Values for Wakeup/Sleep Ratio
For each such defined experiment we have measured various metrics:
— Average Awake Neighbours: this metric represents the average awake neighbours when
the node is in aware of its local neighbourhood, i.e., when the local node is awake. This
average is weighted by the duration of each AwakeNeighbourCount value.
— Scheduled States Distribution: this metric represents the amount of time spent in both
states: wakeup and sleep.
— End-to-End transmission Delay: this metric to show the distribution of the necessary
timespan between the instant when a data packet has been originated at its source, and the
instant when it has been safely received by the sink node. This delay includes all the hops'
transmission duration, but also the time spent in queues across the intermediate nodes
waiting for a path to be set up/repaired.
— Packets Queue Occupancy: this metric represents the distribution of the size of the buffer
queues used to store messages waiting to be sent.
5.4.2 Simulation Results
— Average Awake Neighbours: from figures Figure 64-Figure 65-Figure 66-Figure 67, we can
clearly see that the estimated average awake neighbours computed theatrically thanks to
formulas in (5.10) and (5.19) are confirmed through simulation results.
Figure 64: Average Awake Neighbours in 100 Nodes Network
0
2
4
6
8
10
ρ20% ρ2 ρ3 ρ40% ρ4 ρ60% ρ80%
Avg. AwakeNeighbours
Avg.Neighbours
90
Figure 65: Average Awake Neighbours in 150 Nodes Network
Figure 66: Average Awake Neighbours in 200 Nodes Network
Figure 67: Average Awake Neighbours in 250 Nodes Network
Thus, the formulas in (5.10) and (5.19) are good means in order to compute the target
wakeup/sleep ratio (ρ) in order to have a desired number of the average number of
neighbours. In event-driven applications which use dense network deployment, candidates
for packet forwarding can be numerous. Ideally, 4 neighbours at the four directions can be
used in order to forward packets towards any destination node in the network. This
number can be either reduced to 3 or increased to 4 or other lower/higher values depending
on the distribution of the different potential destination nodes. However, in monitoring
applications, where all communications occur for any node towards the sink node, it is
more important to keep the topology connected by maintaining at least 1 or 2 neighbours
0
2
4
6
8
10
12
ρ2 ρ20% ρ3 ρ4 ρ40% ρ60% ρ80%
Avg. AwakeNeighbours
Avg.Neighbours
0
5
10
15
20
ρ2 ρ3 ρ20% ρ4 ρ40% ρ60% ρ80%
Avg. AwakeNeighbours
Avg.Neighbours
0
5
10
15
20
ρ2 ρ3 ρ20% ρ4 ρ40% ρ60% ρ80%
Avg. AwakeNeighbours
Avg.Neighbours
91
than to allow an easy data packet forwarding in all directions by maintaining a larger
number of neighbours.
— Scheduled States Distribution: Figure 68 shows the distribution of the average sleep
duration according to the different wakeup/sleep ratios and through different topologies
(100, 150, 200 and 250 nodes). In this figure, we can see the chosen ratios (?) result in
different sleep durations. Intuitively, the lower the targeted neighbourhood size, a large
number of nodes switch more frequently and for long periods to the sleep state. Thus, the
overall average sleep duration is higher (ρ2, ρ3, ρ4). Moreover, we can see a correlation
between the percentage of awake neighbours among the total neighbours and the
percentage of the sleep duration among the node lifetime (ρ20%, ρ40%, ρ60%, ρ80%).
Figure 68: Average Sleep Duration (%) of the Network
— End-to-End Delay: Figures Figure 69-Figure 70-Figure 71 represent the average end-to-end
delay for different event-generation average intervals. For this metric, every sensor node
captures data events and reports them to the sink node. Interval between two captured data
event is uniformly distributed with an average of 100s, 150s, and 200s. We can see that
when the average awake neighbourhood size is small (2 awake neighbours), the end-to-end
delay may be higher (~180s). This delay decreases as the awake neighbourhood becomes
more important (less than 50s for neighbourhoods where 80% of the total neighbours are
awake). This increased delay is mainly explained by two factors: (1) long paths, and (2)
queuing delay. Indeed, in topologies where the number of awake neighbours surrounding a
node is smaller, end-to-end path may be rare, and data packets may traverse additional hops
in order to reach the destination node. Moreover, data packets may arrive to an
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
100 Nodes
150 Nodes
200 Nodes
250 Nodes
92
intermediate node with no candidate forwarder. Thus, the packet is stored waiting for a
forwarder to wake up or another route is found.
Figure 69: Average End-to-End Delay (sec) with 100s
Average Event Generation Interval
Figure 70: Average End-to-End Delay (sec) with
150s Average Event Generation Interval
Figure 71: Average End-to-End Delay (sec) with 200s Average Event Generation Interval
End-to-end delay must be carefully taken into account. Indeed, depending on the end
application, a data event can be very urgent to capture and should be rapidly reported. This
can be the case in surveillance applications (battlefield, intrusion-detection, etc.). Other
applications such as natural environment monitoring can tolerate important end-to-end
delays.
— Wait-to-Send Delay: In order to look deep in the end-to-end delay, we have also measured
the wait-to-send delay. This delay represents the amount of time a data packet is kept in the
queue after being received or originated, waiting to be transmitted. The average wait-to-
send delay for different event-generation average intervals is represented in figures Figure
72-Figure 73-Figure 74. We can see that this delay is relatively high, compared to the overall
end-to-end delay. We can also notice that half the end-to-end delay is due is due to time
spent on queues (i.e. wait-to-send delay).
0
50
100
150
200
250
300
N=100
N=150
N=200
N=250
0
50
100
150
200
250
300
N=100
N=150
N=200
N=250
0
50
100
150
200
250
300
ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
N=100
N=150
N=200
N=250
93
Figure 72: Average Wait-to-Send Delay (sec) with 100s
Average Event Generation Interval
Figure 73: Average Wait-to-Send Delay (sec)
with 150s Average Event Generation Interval
Figure 74: Average Wait-to-Send Delay (sec) with 200s Average Event Generation Interval
From the computed delays (end-to-end and wait-to-send), we can see that the larger awake
neighbourhood size, the lower is the end-to-end delay.
— Queue Occupancy: Figures Figure 75-Figure 76-Figure 77 show the average queue
occupancy in different topologies under different data event generations rates. From these
figures, we can estimate the cost of the store-and-forward scheme in terms of memory
space. We can notice that our proposed scheduling and packet forwarding algorithms
achieves resource efficiency since the average number of packets buffered in queues waiting
to be sent do not exceed the value 6 in the highest data rate (i.e. average event interval is
100s). Intuitively, these memory requirements are very low when nodes have more awake
neighbours, and also when the data event rate is lower.
0
20
40
60
80
100
120
140
N=100
N=150
N=200
N=250
0
20
40
60
80
100
120
140
N=100
N=150
N=200
N=250
0
20
40
60
80
100
120
140
ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
N=100
N=150
N=200
N=250
94
Figure 75: Average Queue Occupancy (unit) with 100s
Average Event Generation Interval
Figure 76: Average Queue Occupancy (unit)
with 150s Average Event Generation Interval
Figure 77: Average Queue Occupancy (unit) with 200s Average Event Generation Interval
Since energy conservation and low end-to-end delay are two conflicting objectives,
applications should be carefully designed by finding the best trade-off between the required end-to-
end delay by the application, and the targeted energy conservation.
5.5 Conclusion
In this chapter, we proposed a distributed wakeup/sleep scheduling along packet forwarding
and topology construction/maintenance algorithms. The targeted applications scenarios are
stationary and relatively densely deployed wireless sensor networks used for event-driven data
collection. Such applications include environment monitoring, intrusion detection, etc.
We have considered random WSN deployment, and we have first studied the target
wakeup/sleep ratio targeting a certain number of awake neighbours to be used as forwarders to
relay messages from and through them. Second, we have described an algorithm for topology
construction that is based on minimum hop count towards the destination node (i.e. the sink node).
Finally, we have described a topology maintenance algorithm in order to cope with topology
dynamics, and a data packet forwarding that relies on the store-and-forward scheme.
0,0
5,0
10,0
15,0
20,0
25,0
30,0
35,0
40,0
45,0
ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
N=100
N=150
N=200
N=250
0,0
5,0
10,0
15,0
20,0
25,0
30,0
35,0
ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
N=100
N=150
N=200
N=250
0,0
5,0
10,0
15,0
20,0
25,0
30,0
ρ2 ρ3 ρ4 ρ20% ρ40% ρ60% ρ80%
N=100
N=150
N=200
N=250
95
Simulations results confirmed the theoretical results and showed that the proposed
mechanisms can aid the application design by allowing finding the best trade-off between
application requirements (end-to-end delay and buffer space) and the targeted energy efficiency
through the parameterization of the wakeup/sleep ratio.
96
Chapter 6
6 Conclusion and Perspectives
This chapter concludes the thesis and summarizes the major contributions while
highlighting future research direction sand perspectives.
Challenged multihop wireless networks are often battery-powered. Thus, communications
protocols must be energy efficient in order to maximize the network lifetime and so, the
application lifetime. Moreover, due to low transmit power, wireless links are subject to various
issues such as short range, radio interference, prone to errors and loss, etc. making the wireless links
between the communicating nodes very unreliable. Furthermore, and depending on the application
scenario, nodes may be mobile or subject to failure due to energy depletion, physical damage, theft,
etc. resulting in a dynamic network topology. Along these stringent constraints, applications have
demanding requirements such as low end-to-end delays, high throughput, high fault-tolerance, etc.
In this thesis, and considering the aforementioned issues, we have proposed efficient routing
protocols for challenged multihop wireless networks satisfying different application's requirements
while achieving resource efficiency. We believe that efficiency can only be achieved through fully
distributed algorithms tending to zero control packets.
The summary of the thesis contributions is presented below.
6.1 Thesis Contributions
In this thesis, we started by studying the different characteristics of challenged multihop
wireless networks, but also the principles and the open issues that govern the design of application
using such networks.
As a first challenge, we have tackled the problem of bandwidth demanding applications that
make use of wireless multimedia sensor networks (WMSNs). WMSNs are more and more used for
video surveillance applications, intrusion detection, environment monitoring, etc. These
applications have high requirements in terms of throughput, delay, packet loss, etc. such
requirements can be bigger than nodes capabilities. Thus, distributed techniques must be employed
in order to guarantee the service. For such applications, multipath routing might be appropriate
since it allows parallel transmissions of data sub-streams that will be re-gathered at the destination
node. However, the utilization of multiple paths may introduce extra overhead by generating
97
consequent control traffic in the network in order to explore, construct and maintain the multiple
paths. In this context, we proposed an online multipath routing protocol (AGEM) that transmits
multimedia streams through multiple paths established in an online fashion and without topology
exploration. AGEM relies on greedy geographic packet forwarding. It ensures uniform energy
consumption among the nodes of the network in order to maximize the network lifetime, and
meets the applications requirements through low end-to-end delay, low packet loss ratio, and
increased throughput as it has been proved through extensive simulations against single path and
offline multipath routing protocols.
Second, we have tackled the problem of routing in delay tolerant networks (DTNs). DTNs
can be very challenging since they suffer from intermittent connectivity. At any instant, a path
between a source node and the destination may not necessarily exist. Nodes rely on multihop
communications and use store-carry-and-forward scheme for packet forwarding. Applications
developed for DTNs must tolerate higher delays, frequent disconnections, etc. but aim to
communicate with bounded delays and by using efficiently the network resources. In this context,
we have proposed an efficient routing protocol for city-wide data dissemination and collection,
namely ORION. After studying the inter-node contact behavior in the case of a city-wide mobile
network composed of different types of nodes (regular, irregular and static), we have derived an
inter-node contact model. This model is at the center of ORION protocol, since it is used to
forecast future contact dates and durations between the network nodes. Based on this information
and other information such as geographic coordinates, trajectory assistance and greedy forwarding,
ORION protocol performs its routing decisions. Through simulations, we have shown that
ORION protocol outperforms existing protocols and achieves good performances in terms of
packet delivery ratio, end-to-end delay, but also resources efficiency (memory space for packet
queues).
Finally, we have tackled the problem of energy-efficient routing in stationary wireless
sensor networks (WSNs) for event-driven applications. For such application and depending on the
specified event frequency, some nodes can be put on a low energy-consuming state in order to
maximize the network lifetime. In this context, we started by studying the distributed
wakeup/sleep scheduling based on statistical distributions. We have shown theoretically, that the
best schedule can be found in order to achieve a desired network topology property, represented by
the average awake neighbors. Through path construction and maintenance and packet forwarding
mechanisms, we have confirmed by extensive simulations our theoretical analysis. Simulation
results have also shown good performances in terms of end-to-end delay, energy consumption and
resources utilization (memory space for packet queues). Based on the obtained results, application
98
developer can derive the values of the different parameters in order to achieve a certain tradeoff
between applications requirements and network resources constraints.
6.2 Perspectives
During the thesis, we have faced various challenges. Future research directions can be
summarized as follows:
— Localization Error-Tolerance: a lot of routing protocols are assuming that nodes have the
knowledge of their geographic location. Based on this information, geographical routing
protocols, that make use of greedy packet forwarding, have been proposed. However, in
practice, it is rarely possible to embed localization hardware within the wireless node
module. Thus, nodes make use of distributed localization techniques in order to have an
approximate location computed through various algorithms. These obtained locations are
not error-free. Thus, routing decisions based on inaccurate information may be inefficient
and may lead to energy depletion. A future direction to our works is to include
localization-error into account in order to optimize the resources and to achieve good
routing performances.
— Contact model learning and Adaptability: in our routing protocol for delay tolerant
networks, the inter-node contact model has been learned and derived offline prior to
application deployment. The contact model has been retained as it was empirically
dominant. A future direction to this work is to make the nodes capable of learning the
contact-model in an online fashion, but also to adapt in case where a learned contact
model does not apply anymore.
— Adaptive duty-cycle algorithm: in our topology control, the scheduling parameters were
computed offline and applied to the entire network in order to optimize the overall
network performances. However, it would be more efficient if wireless nodes could adapt
their wakeup/sleep schedule according to their local neighborhood in space (according to
the resulting deployment) and time (according to neighbors' schedules).
— Implementation of a proof of concept prototype: all the proposed mechanisms in this
thesis were proven and evaluated through either theoretical analysis or through extensive
simulations. Although, resource utilization such as memory and processing power has been
taken into account in all our contributions' design steps, these means are insufficient in
order to prove the real performances. Therefore, we plan to implement the proposed
protocols into real sensor nodes composed on Oracle SPOTs (java powered sensor nodes).
99
Publications
Journals
Samir Medjiah, Toufik Ahmed and Abolghasem (Hamid) Asgari "Streaming Multimedia over
WMSNs: An Online Multipath Routing Protocol", In InderScience International Journal on Sensor
Networks (IJSNet), Special Issue on Multimedia Data Applications in Wireless Sensor Networks
(InderScience IJSNet).
International Conferences
Samir Medjiah and Toufik Ahmed, "ORION Routing Protocol for Delay Tolerant Networks", In
the 46th Annual IEEE International Conference on Communications 2011 (IEEE ICC 2011). June, 5th
- 9th, 2011, Kyoto, Japan.
Samir Medjiah, Toufik Ahmed and Francine Krief, "AGEM: an Adaptive Greedy-compass Energy-
aware Multipath Routing Protocol for WMSNs", in The 7th Annual IEEE Consumer
Communications and Networking Conference 2010 (IEEE CCNC 2010). January, 9th - 12th, 2010,
Las Vegas, USA.
Samir Medjiah, Toufik Ahmed and Francine Krief, "GEAMS: a Greedy Energy-Aware Multipath
Stream-based Routing Protocol for WMSNs", in the 2nd IEEE Global Information Infrastructure
Symposium (IEEE GIIS 2009), 23rd - 26th, June, 2009, Hammamet, Tunisia.
National Conferences
Samir Medjiah, Toufik Ahmed, Francine Krief et Patrick Gélard, "AGEM: un Protocole de Routage
Géographique Angulaire Adaptatif", Colloque Francophone sur l'Ingénierie des Protocoles 2009 (CFIP
2009). 12-15 Octobre 2009, Strasbourg, France.
Workshops
Patrick Gélard, Charles Yana, Emmanuel Dubois, Samir Medjiah, Toufik Ahmed and Francine
Krief, "Hybridization Architecture of Satellite Networks with Wireless Sensor Networks", in The
ESA Wireless Sensor Networks for Space Applications Workshop 2009 (ESA WiSens4Space 2009), 1st -
2nd October, 2009, Santorini, Greece.
100
References
[1] Gad-el-Hak M (ed) 2002 The MEMS Handbook (Boca Raton, FL: CRC Press)
[2] Akyildiz, I., Wang, X., and Wang, W. (2005). Wireless mesh networks: a survey. Computer Networks,
47(4):445–487.
[3] Potdar, V.; Sharif, A.; Chang, E.; , "Wireless Sensor Networks: A Survey," Advanced Information
Networking and Applications Workshops, 2009. WAINA '09. International Conference on , vol., no., pp.636-
641, 26-29 May 2009
[4] I.F. Akyildiz et al., Wireless sensor networks: a survey, Computer Networks 38 (4) (2002) 393–422.
[5] Watteyne, T.; Molinaro, A.; Richichi, M.G.; Dohler, M.; , "From MANET To IETF ROLL
Standardization: A Paradigm Shift in WSN Routing Protocols," Communications Surveys & Tutorials, IEEE ,
vol.13, no.4, pp.688-707, Fourth Quarter 2011
[6] I.F. Akyildiz, T. Melodia, K.R. Chowdhury, A survey on wireless multimedia sensor networks, Computer
Networks 51 (4) (2007) 921–960.
[7] I.F. Akyildiz, I.H. Kasimoglu, Wireless sensor and actor networks: research challenges, Ad Hoc Networks
Journal 2 (4) (2004) 351–367.
[8] I.F. Akyildiz, D. Pompili, T. Melodia, Underwater acoustic sensor networks: research challenges, Ad Hoc
Networks 3 (3) (2005) 257–279.
[9] Munir, S.A., Ren, B., Jiao, W., Wang, B., Xie, D., Ma, J.: Mobile wireless sensor network: Architecture and
enabling technologies for ubiquitous computing. In: Proceedings of the 21st International Conference on
Advanced Information Networking and Applications Workshops, AINAW (2007)
[10] K. Harras, K. Almeroth, and E. Belding-Royer, “Delay tolerant mobile networks (DTMNs): Controlled
flooding schemes in sparse mobile networks,” in Proc. Networking, 2005.
[11] B. T. Clough, “Unmanned aerial vehicles: autonomous control challenges, a researcher’s perspective,” in
Cooperative Control and Optimization, R.Murphey and P. M. Pardalos, Eds. Norwell, MA: Kluwer, 2000,
pp. 35–53.
[12] A . Arora, P. Dutta, S. Bapat, V. Kulathumani, H. Zhang, V. Naik, V. Mittal, H. Cao, M. Demirbas, and M.
Gouda, “A line in the sand : A wireless sensor network for target detection, classification, and tracking,”
Computer Networks, vol. 46, no. 5, pp. 605–634, 2004.
[13] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler, and J. Anderson, “Wireless sensor networks for habitat
monitoring,” in ACM WSNA, 2002, pp. 88–97.
[14] S. Kim, “Wireless Sensor Networks for Structural Health Monitoring,” Master’s thesis, University of
California at Berkeley, March 2005.
[15] K. Kim, J. Jun, S. Kim, and B. Sung, “Medical asset tracking application with wireless sensor networks,” in
SENSORCOMM, 2008, pp. 531–536.
[16] S. Nourizadeh, C. Deroussent, Y. Song, and J. Thomesse, “Medical and home automation sensor networks
for senior citizens telehomecare,” in ICC Communications Workshops, 2009, pp. 1–5.
[17] R. Aylward and J. Paradiso, “A compact, high-speed, wearable sensor network for biomotion capture and
interactive media,” in Proceedings of IPSN, 2007, pp. 380–389.
[18] N. Ota and P. Wright, “Trends in wireless sensor networks for manufacturing,” Int. J. Manufacturing
Research, vol. 1, no. 1, pp. 3–17, 2006.
101
[19] D.Malan, T. Fulford-Jones,M.Welsh, and S.Moulton, “Codeblue : an ad hoc sensor network infrastructure
for emergency medical care,” in Int. Workshop Wearable Implantable Body Sens. Networks, Apr. 2004.
[20] H. Karl and A. Willig, Protocols and architectures for wireless sensor networks. John Wiley & Sons Ltd,
2005.
[21] Kavi K. Khedo, Rajiv Perseedoss, Avinash Mungu, “A Wireless Sensor Network Air Pollution Monitoring
System”, International Journal Of Wireless & Mobile Network (IJWMN) 2010, Vol. 2, No.2, P 31, ISSN
0975 -3834.
[22] F. Koushanfar, M. Potkonjak, and A. Sangiovanni-Vincentelli, “Fault Tolerance in Wireless Ad hoc Sensor
Networks”. Proceedings of IEEE Sensors 2002, June 2002.
[23] L. Paradis and Q. Han, “A Survey of Fault Management in Wireless Sensor Networks”. Plenum Press New
York, NY, USA, 2007.
[24] F. Nekoogar, F. Dowla, and A. Spiridon, “Self organization of wireless sensor networks using ultra-
wideband radios”. Atlanta, GA, United States, September 2004.
[25] V. Handziski, A. Kopke, H. Karl, and A. Wolisz, “A common wireless sensor network architecture”.
Technische Universität Berlin, July 2003, pp.10-17.
[26] V. Handziski, J. Polastrey, J. H. Hauer, C. Sharpy, A. Wolisz, and D. Cullery, “Flexible Hardware
Abstraction for Wireless Sensor Networks”. In Proceedings of the Second European Workshop on Wireless
Sensor Networks (EWSN '05),February 2005.
[27] C.Y. Chong and S.P. Kumar, “Sensor networks: Evolution, opportunities, and challenges”. Proceedings of
the IEEE, vol. 91, n.8, 2003, pp. 1247-1256.
[28] T. Zhao, W. D. Cai, and Y. J. Li, “A Sensor Network Topology Inference Algorithm, omputational
Intelligence and Security”. 2007 International Conference on. January 2008.
[29] I. Teixeira, J. F. de Rezende, A. de Castro, and A. C. P. Pedroza, “Wireless Sensor Network: Improving the
Network Energy Consumption”. in XXI Symposium Brazilian Telecommunications, SBT’04, Belem, Brazil,
September 2004.
[30] Tilak, S., Abu-Ghazaleh, N., and Heinzelman, W. (2002). A taxonomy of wireless microsensor network
models. ACM SIGMOBILE Mobile Computing and Communications Review, 6(2):28–36.
[31] Chebrolu, K., Raman, B., Mishra, N., Valiveti, P. K., and Kumar, R. (2008). Brimon: a sensor network
system for railway bridge monitoring. In Proceedings of the sixth International Conference on Mobile
System, Applications, and Services (MobiSys’08).
[32] M. Ali and S. K. Ravula, “Real-time support and energy efficiency in wireless sensor networks”. Technical
report, IDE0805, January 2008.
[33] Jamal N. Al-Karaki and Ahmed E. Kamal, "Routing techniques in wireless sensor networks: a survey," IEEE
Wireless Communications, vol. 11, no. 6, pp. 6-28, 2004.
[34] C. Intanagonwiwat, R. Govindan, and D. Estrin, "Directed diffusion: a scalable and robust communication
paradigm for sensor networks," Proceedings of ACM MobiCom '00, Boston, MA, 2000, pp. 56-67.
[35] W. Heinzelman, J. Kulik, and H. Balakrishnan, "Adaptive Protocols for Information Dissemination in
Wireless Sensor Networks," Proc. 5th ACM/IEEE Mobicom Conference (MobiCom '99), Seattle, WA,
August, 1999. pp. 174-85.
[36] D. Braginsky and D. Estrin, "Rumor Routing Algorithm for Sensor Networks," in the Proceedings of the
First Workshop on Sensor Networks and Applications (WSNA), Atlanta, GA, October 2002.
102
[37] M. Chu, H. Haussecker, and F. Zhao, "Scalable Information-Driven Sensor Querying and Routing for ad
hoc Heterogeneous Sensor Networks," The International Journal of High Performance Computing
Applications, Vol. 16, No. 3, August 2002.
[38] Y. Yao and J. Gehrke, "The cougar approach to in-network query processing in sensor networks", in
SIGMOD Record, September 2002.
[39] N. Sadagopan et al., The ACQUIRE mechanism for efficient querying in sensor networks, in the
Proceedings of the First International Workshop on Sensor Network Protocol and Applications,
Anchorage, Alaska, May 2003.
[40] W. Heinzelman, A. Chandrakasan and H. Balakrishnan, "Energy-Efficient Communication Protocol for
Wireless Microcrosensor Networks," Proceedings of the 33rd Hawaii International Conference on System
Sciences (HICSS '00), January 2000.
[41] S. Lindsey, C. Raghavendra, "PEGASIS: Power-efficient Gathering in Sensor Information Systems", IEEE
Aerospace Conference Proceedings, 2002, Vol. 3, 9-16 pp. 1125-1130.
[42] A. Manjeshwar and D. P. Agarwal, "TEEN: a routing protocol for enhanced efficiency in wireless sensor
networks," In 1st International Workshop on Parallel and Distributed Computing Issues in Wireless
Networks and Mobile Computing, April 2001.
[43] A. Manjeshwar and D. P. Agarwal, "APTEEN: A hybrid protocol for efficient routing and comprehensive
information retrieval in wireless sensor networks," Parallel and Distributed Processing Symposium.,
Proceedings International, IPDPS 2002, pp. 195-202.
[44] V. Rodoplu and T. H. Meng,"Minimum Energy Mobile Wireless Networks", IEEE Journal Selected Areas
in Communications, vol. 17, no. 8, Aug. 1999, pp. 133344.
[45] Q. Fang, F. Zhao, and L. Guibas, "Lightweight Sensing and Communication Protocols for Target
Enumeration and Aggregation", Proceedings of the 4th ACM international symposium on Mobile ad hoc
networking and computing (MOBIHOC), 2003, pp. 165-176.
[46] Jamal N. Al-Karaki, Raza Ul-Mustafa, Ahmed E. Kamal, "Data Aggregation in Wireless Sensor Networks -
Exact and Approximate Algorithms'", Proceedings of IEEE Workshop on High Performance Switching and
Routing (HPSR) 2004,April 18-21, 2004, Phoenix, Arizona, USA.
[47] Q. Li and J. Aslam and D. Rus,"Hierarchical Power-aware Routing in Sensor Networks", In Proceedings of
the DIMACS Workshop on Pervasive Networking, May, 2001.
[48] F. Ye, H. Luo, J. Cheng, S. Lu, L. Zhang, "A Two-tier data dissemination model for large-scale wireless
sensor networks",proceedings of ACM/IEEE MOBICOM, 2002.
[49] B. Karp and H. T. Kung, "GPSR: Greedy perimeter stateless routing for wireless sensor networks", in the
Proceedings of the 6th Annual ACM/IEEE International Conference on Mobile Computing and
Networking (MobiCom '00), Boston, MA, August 2000.
[50] Y. Xu, J. Heidemann, D. Estrin,"Geography-informed Energy Conservation for Ad-hoc Routing," In
Proceedings of the Seventh Annual ACM/IEEE International Conference on Mobile Computing and
Networking 2001, pp. 70-84.
[51] Y. Yu, D. Estrin, and R. Govindan, "Geographical and Energy-Aware Routing: A Recursive Data
Dissemination Protocol for Wireless Sensor Networks", UCLA Computer Science Department Technical
Report, UCLA-CSD TR-01-0023, May 2001.
103
[52] I. Stojmenovic and X. Lin. "GEDIR: Loop-Free Location Based Routing in Wireless Networks", In
International Conference on Parallel and Distributed Computing and Systems, Boston, MA, USA, Nov. 3-6,
1999.
[53] F. Kuhn, R. Wattenhofer, A. Zollinger,"Worst-Case optimal and average-case efficient geometric ad-hoc
routing", Proceedings of the 4th ACM International Conference on Mobile Computing and Networking,
Pages: 267-278, 2003.
[54] B. Chen, K. Jamieson, H. Balakrishnan, R. Morris, "SPAN: an energy-efficient coordination algorithm for
topology maintenance in ad hoc wireless networks", Wireless Networks, Vol. 8, No. 5, Page(s): 481-494,
September 2002.
[55] L. Shu, Y. Zhang, L.T. Yang, Y. Wang, M. Hauswirth, N. Xiong, TPGF: geographic routing in wireless
multimedia sensor networks, Telecommunication Systems November 05 (2009), doi: 10.1007/s11235-009-
9227-0.
[56] L. Zhang, M. Hauswirth, L. Shu, Z. Zhou, V. Reynolds and G. Han, "Multi-priority Multi-path Selection
for Video Streaming in Wireless Multimedia Sensor Networks", Lecture Notes in Computer Science, vol.
5061, p.439-452. 2008.
[57] K. Sohrabi, J. Pottie, "Protocols for self-organization of a wireless sensor network", IEEE Personal
Communications, Volume 7, Issue 5, pp 16-27, 2000.
[58] T. He et al., "SPEED: A stateless protocol for real-time communication in sensor networks", in the
Proceedings of International Conference on Distributed Computing Systems, Providence, RI, May 2003.
[59] T. Spyropoulos, K. Psounis, and C. S. Raghavendra “Single-copy Routing in Intermittently Connected
Mobile Networks,” First Annual IEEE Communications Society Conference on Sensor and Ad Hoc
Communications and Networks, 2004, pp.235-244.
[60] Keranen, A., OTT, J., and Karkkainen, (2009), T. The ONEsimulator for DTN protocol evaluation. In
Simutools ’09: Proceedings of the 2nd International Conference on Simulation Tools and Techniques ,ICST
(Institute for Computer Sciences, Social- Informatics and Telecommunications Engineering), pp. 1–10
[61] C.E. Perkins and E.M. Royer, “Ad-Hoc On Demand Distance Vector Routing,” Proceedings of the 2nd IEEE
Workshop on Mobile Computing Systems and Applications (WMCSA), New Orleans, LA, February 1999, pp.
90-100.
[62] D. B. Johnson, and D. A. Maltz, “Dynamic Source Routing in Ad Hoc Wireless Network,” T. Imielinkski
and H. Korth, Eds., Mobile Computing. Kluwer Academy, 1996, pp. 153–181.
[63] U. Javaid, DE Meddour, MR Tinku, T Ahmed, "Towards Universal Convergence in Heterogeneous
Wireless Networks using Ad Hoc Connectivity," 9th Intl. Symposium on Wireless Personal Multimedia
Communications (WPMC'06), 2006
[64] Melissa Ho, Kevin Fall: Delay Tolerant Networking for Sensor Networks, in Proceedings of the IEEE
International Conference on Sensing, Communications and Networking (IEEE SECON) 2004.
[65] Pelusi, L.; Passarella, A.; Conti, M.; , "Opportunistic networking: data forwarding in disconnected mobile
ad hoc networks," Communications Magazine, IEEE , vol.44, no.11, pp.134-141, November 2006
[66] Vahdat, A., Becker, D.: Epidemic Routing for Partially Connected Ad Hoc Networks. Technical Report, CS-
2000-06, Duke University (2000)
[67] Cong Liu; Jie Wu; "Efficient Adaptive Routing in Delay Tolerant Networks," Communications, 2009. ICC
'09. IEEE International Conference on, vol., no., pp.1-5, 14-18 June 2009.
104
[68] Wei Gao; Guohong Cao; , "On exploiting transient contact patterns for data forwarding in Delay Tolerant
Networks," Network Protocols (ICNP), 2010 18th IEEE International Conference on , vol., no., pp.193-202, 5-
8 Oct. 2010
[69] Kevin Fall, Keith L. Scott, Scott C. Burleigh, Leigh Torgerson, Adrian J.Hooke, Howard S. Weiss, Robert
C. Durst, and Vint Cerf, Delay-Tolerant Networking Architecture, http://tools.ietf.org/html/rfc4838
(2007).
[70] Kate, A., Zaverucha, G. M. & Hengartner, U. (2007). Anonymity and security in delay tolerant networks,
Security and Privacy in Communications Networks and the Workshops, 2007. SecureComm 2007. Third
International Conference on, pp. 504–513.
[71] Jieyan Liu, Haigang Gong, and Jiazhi Zeng. Preference location-based routing in delay tolerant networks.
International Journal of Digital Content Technology and its Applications, 5(12):468–474, December 2011.
[72] Cong Liu, Jie Wu, Scalable routing in delay Tolerant Networks: in Proceedings of Mobihoc’07 the 8th ACM
international symposium on Mobile ad hoc networking and computing. pp. 51–60, 2007
[73] M. Liu, Y. Yang, and Z. Qin, “A survey of routing protocols and simulations in Delay-Tolerant Networks,”
in Proc. of WASA, 2011, pp. 243–253.
[74] Fall, K. "A Delay-Tolerant Network Architecture for Challenged Internets. In: Proc. of the ACM
SIGCOMM, pp. 27–34. ACM Press, Karlsruhe (2003)
[75] Chen, X., Murphy, A.L.: Enabling Disconnected Transitive Communication in Mobile Ad Hoc Networks.
In: Proceedings of ACM POMC (2001)
[76] Dubois-Ferriere, H., Grossglauser, M., Vetterli, M.: Age Matters: Efficient Route Discovery in Mobile Ad
Hoc Networks using Encounter Ages. In: Proceedings of ACM MobiHoc (2003)
[77] Yuan, Q., Cardei, I., Wu, J.: Predict, Relay. An Efficient Routing in Disruption-Tolerant Networks. In:
MobiHoc 2009, pp. 95–104 (2009)
[78] Daly, E., Haahr, M.: Social Network Analysis for Routing in Disconnected Delay-Tolerant MANETs. In:
Proc. of the MobiHoc 2007, pp. 32–40 (2007)
[79] Li, Q., Zhu, S., Cao, G.: Routing in Socially Selfish Delay Tolerant Networks. In: Proceedings of IEEE
INFOCOM 2010, pp. 1–9 (March 2010).
[80] Spyropoulos, T., Psounis, K., and Raghavendra, C. S. 2005. “Spray and wait: an efficient routing scheme for
intermittently connected mobile networks.” In Proceedings of the 2005 ACM SIGCOMM Workshop on Delay-
Tolerant Networking (Philadelphia, Pennsylvania, USA, August 26 - 26, 2005). WDTN '05. ACM, New
York, NY, 252-259.
[81] T. Spyropoulos, K. Psounis, and C. Raghavendra. “Spray and Focus: Efficient Mobility-Assisted Routing for
Heterogeneous and Correlated Mobility.” In Proc. of IEEE PerCom, 2007.
[82] Bulut, E., Wang, Z., Szymanski, B.K.: Cost-Effective Multiperiod Spraying for Routing in Delay-Tolerant
Networks. TON, 1530–1543 (2010)
[83] Pan, H., Crowcroft, J.E.Y.: BUBBLE rap: Social-Based forwarding in delay tolerant networks. In: Proc. of
the MobiHoc 2008, pp. 241–250. ACM Press, Hong Kong (2008)
[84] Costa, P., Mascolo, C., Musolesi, M., Picco, G.P.: Socially-Aware routing for publishsubscribe in delay-
tolerant mobile ad hoc networks. IEEE Journal of Selected Areas in Communication 26(5), 748–760 (2008)
[85] Balasubramanian, A., Levine, B.N., Venkataramani, A.: DTN routing as a resource allocation problem. In:
Proc. ACM SIGCOMM (2007)
105
[86] Jiang, Y., Li, Y., Zhou, L.: Optimal Opportunistic Forwarding with Energy Constraint for DTN. In: 2010
INFOCOM IEEE Conference on Computer Communications Workshops, pp. 1–2 (March 2010)
[87] Lin, Y.F., Li, B.C., Liang, B.: Stochastic analysis of network coding in epidemic routing. IEEE Journal on
Selected Areas in Communications 26(5), 794–808 (2008)
[88] Widmer, J., Boudec, J.L.: Network coding for efficient communication in extreme networks. In: Proc. of the
ACM SIGCOMM Workshop on DTN, pp. 284–291 (2005)
[89] Lee, K., Yi, Y., Jeong, J.: Max-Contribution: On Optimal Resource Allocation in Delay Tolerant
Networks. In: Proceedings of IEEE INFOCOM 2010, pp. 1–9 (March 2010)
[90] Whitbeck, J., Conan, V.: HYMAD: Hybrid DTN-MANET routing for dense and highly dynamic wireless
networks. Computer Communications 33(13), 1483–1492 (2010).
[91] G. Toussaint. The relative neighbourhood graph of a finite planar set. Pattern Recognition, 12 :261–268,
1980.
[92] Prosenjit Bose, Pat Morin, Ivan Stojmenovic, and Jorge Urrutia. Routing with guaranteed delivery in ad hoc
wireless networks. Wireless Networks, 7(6) :609–616, November 2001.
[93] Prosenjit Bose, Luc Devroye, William Evans, and David Kirkpatrick. On the spanning ratio of gabriel
graphs and beta-skeletons. SIAM Journal on Discrete Mathematics, 20(2) :412–427, February 2006.
[94] Rajmohan Rajaraman. Topology control and routing in ad hoc networks: a survey. ACM SIGACT News,
33(2) :60–73, June 2002
[95] Julien Cartigny, Francois Ingelrest, David Simplot-Ryl, and Ivan Stojmenovic. Localized LMST and rng
based minimum-energy broadcast protocols in ad hoc networks. Ad Hoc Networks, 3(1) :1–16, November
2005
[96] Ruben K. Gabriel and Robert R. Sokal. A new statistical approach to geographic variation analysis.
Systematic Zoology, 18(3) :259–278, September 1969.
[97] Ning Li, Hou J.C., and Sha L. Design and analysis of an MST-based topology control algorithm. IEEE
Transactions on Wireless Communications, 4(3):1195–1206, May 2005.
[98] R. C. Prim. Shortest connection networks and some generalizations. Bell Systems Technical Journal, 1 :1389–
1401, November 1957.
[99] J. Wu and F. Dai. Distributed dominant pruning in ad hoc wireless networks. In 38th annual IEEE
International Conference on Communications (ICC), Anchorage, USA, May 2003.
[100] A. Qayyum, L. Viennot, and A. Laouiti. Relaying for flooding broadcast messages in mobiles wireless
networks. In 35th annual Hawaii international conference on system sciences (HICSS), Hawaii, USA, January
2002.
[101] T. Clausen, P. Jacquet, C. Adjih, A. Laouiti, P. Minet, P. Muhlethaler, A. Qayyum, and L. Viennot.
Optimized link state routing protocol (OLSR). RFC 3626 (Experimental), October 2003.
[102] C. Adjih, P. Jacquet, and L. Viennot. Computing CDS with multipoint relays. Technical Report RR-4597,
INRIA, October 2002.
[103] I. Amadou and F. Valois, “Performance Evaluation of Distributed Self-Organization Protocols in Wireless
Sensor Networks,” in 7th ACM International Symposium on Performance Evaluation of Wireless Ad Hoc,
Sensor, and Ubiquitous Networks (PE-WASUN), Bodrum, Turquie, Oct. 2010.
[104] P. Wan, K. Alzoubi, and O. Frieder. Distributed construction of connected dominating set in wireless ad
hoc networks. In 21st Annual Joint Conference of the IEEE Computer and Communications Societies
(INFOCOM), New York, USA, June 2002.
106
[105] Gurses, E. and Akan, O.B. (2005) ‘Multimedia Communication in Wireless Sensor Networks’, Annals of
Telecommunications, Vol. 60, No. 7-8, pp. 799-827.
[106] Misra, S., Reisslein, M. and Guoliang, X. (2008) ‘A Survey of Multimedia Streaming in Wireless Sensor
Networks’, IEEE Communications Surveys & Tutorials, Vol. 10, No. 4, pp. 18-39.
[107] Shu, L. and Chen, M. (2010) ‘Multimedia over Sensor Networks’, IEEE ComSoc MMTC E-Letter, Vol. 5, No.
5, pp 59-61.
[108] Aivaloglou, E., Gritzali, S. and Skianis, C. (2008) ‘Trust Establishment in Sensor Networks: Behavior-based,
Certificate-based, and a Combinational Approach’, International Journal of System of Systems Engineering
(IJSSE) 1 (1-2), pp. 128-148.
[109] Vassis, D., Kormentzas, G. and Skianis, C. (2006) ‘Performances Evaluation of Single and Multi-Channel
Actor to Actor Communication for Wireless Sensor Actor Networks’, Ad Hoc Networks 4 (4), Elsevier
Science, pp. 487-498.
[110] T. Ahmed, R. Boutaba, A. Mehaoua, A measurement-based approach for dynamic QoS adaptation in
DiffServ networks, Journal of Computer Communications (2004) (special issue on End-to-End Quality of
Service Differentiation)
[111] E. Borcoci, A. Asgari, N. Butler, T. Ahmed, A. Mehaoua, G. Kormentzas, S. Eccles, Service management
for end-to-end QoS multimedia content delivery in heterogeneous environment, in: Proceedings of AICT
Conference, July 2005, Lisbon
[112] A. Nafaa, T. Ahmed, A. Mehaoua, Unequal and interleaved FEC protocol for robust MPEG-4 multicasting
over wireless LANs, in: Proceedings of IEEE ICC'04, The 2004 IEEE International Conference on
Communications, Paris, France, July 2004, vol. 3, pp. 1431–1435
[113] I. Djama and T. Ahmed, “A cross-layer interworking of DVB-T and WLAN for mobile IPTV service
delivery,” IEEE Trans. Broadcasting, vol. 53, no. 1, pp. 382–390, Mar. 2007
[114] Li, S., Kisore Neelisetti, R., Liu, C. and Lim, A. (2010) ‘Efficient Multi-path Protocol for Wireless Sensor
Networks’, International Journal of Wireless & Mobile Networks, Vol. 2, No. 1, pp. 110-130.
[115] Vidhyapriya, R. and Vanathi, P.T. (2007) ‘Energy Efficient Adaptive Multipath Routing for Wireless Sensor
Networks’, IAENG International Journal of Computer Science 34 (1).
[116] Maimour, M. (2008) ‘Maximally Radio-Disjoint Multipath Routing for Wireless Multimedia Sensor
Networks’, in Proceedings of WMuNep 2008.
[117] Huang, X. and Fang, Y. (2008) ‘Multiconstrained QoS Multi-path Routing in Wireless Sensor Networks’,
Wireless Networks, Vol. 14, No. 4, pp. 465-478.
[118] Stojmenovic, I. and Lin, X. (1999) ‘GEDIR: Loop Free Location Based Routing in Wireless Communication
Network’, Proceedings of IASTED International Conference on Parallel and Distributed Computing Systems, pp.
1025-1028.
[119] Li, J., Jannotti, J., DeCouto, D. S. J., Karger, D. R. and Morris, R. (2000) ‘A Scalable Location Service for
Geographic Ad Hoc Routing’, Proceedings of ACM/IEEE MobiCom.
107
[120] Stojmenovic, I. (2002) ‘Position-based Routing in Ad Hoc Networks’, IEEE Communication Magazine, Vol.
40, pp. 128–134.
[121] Kranakis, E., Singh, H. and Urrutia, J. (1999) ‘Compass Routing on Geometric Networks’, Proceedings of the
11th Canadian Conference on Computational Geometry.
[122] Morin, P. R. (2001) ‘On line routing in geometric graphs’, Ph.D. Thesis, School of Computer Science,
Carleton University.
[123] Urrutia, J. (2002) ‘Routing with guaranteed delivery in geometric and Wireless Networks’, Handbook of
Wireless Networks and Mobile Computing, pp. 393–406.
[124] Stojmenovic, I. and Lin, X. (2001) ‘Loop-free Hybrid Single Path/Flooding Routing Algorithms with
Guaranteed Delivery for Wireless Networks’, IEEE Transactions on Parallel Distributed Systems, 12(10).
[125] Bose, P. and Morin, P. (1999) ‘Online Routing in Triangulations’, Proceedings of the 10th Annual
International Symposium on Algorithms Computation ISAAC.
[126] Medjiah, S.; Ahmed, T.; Krief, F.; , "AGEM: Adaptive Greedy-Compass Energy-Aware Multipath Routing
Protocol for WMSNs," Consumer Communications and Networking Conference (CCNC), 2010 7th IEEE ,
vol., no., pp.1-6, 9-12 Jan. 2010.
[127] Medjiah, S., Ahmed, T. and Asgari, A.H, “Streaming Multimedia over WMSNs: An Online Multipath
Routing Protocol”, International Journal of Sensor Networks (IJSNet), vol. 11, no. 1, pp 10-21, 2012.
[128] Niculescu, D. and Nath, B. (2003) ‘DV Based Positioning in Ad hoc Networks’, In Journal of
Telecommunication Systems.
[129] Nagpal, R. Shrobe, H. and Bachrach, J. (2003) ‘Organizing a Global Coordinate System from Local
Information on an Ad Hoc Sensor Network’, Proceedings of the 2nd International Workshop on Information
Processing in Sensor Networks (IPSN '03).
[130] Heinzelman, W., Chandrakasan, A. and Balakrishnan, H. (2000) ‘Energy-Efficient Communication Protocol
for Wireless Microsensor Networks’, Proceedings of the 33rd Hawaii International Conference on System
Sciences (HICSS), pp. 1–10. 4–7.
[131] Varga, A. and Hornig, R. (2008) ‘An overview of the OMNeT++ simulation environment’, Proceedings of
the 1st International Conference on Simulation Tools and Techniques for Communications, Networks and
Systems (Simutools '08).
[132] Medjiah, S.; Ahmed, T.; Krief, F.; , "GEAMS: A Geographic Energy-Aware Multipath Stream-based routing
protocol for WMSNs," Information Infrastructure Symposium, 2009. GIIS '09. Global , vol., no., pp.1-8, 23-
26 June 2009
[133] U. Javaid, F. Rasheed, D.-E. Meddour & T. Ahmed. (2008) “Adaptive distributed gateway discovery in
hybrid wireless networks.” In IEEE Wireless Communications and Networking Conference. pp 2735-2740
[134] The OpenMesh Project, Internet link : http://www.openmeshproject.org/
[135] Medjiah, S.; Ahmed, T.; , "Orion Routing Protocol for Delay Tolerant Networks," Communications (ICC),
2011 IEEE International Conference on , vol., no., pp.1-6, 5-9 June 2011
[136] A. Lindgren, A. Doria, and O. Schelen. “Probabilistic Routing in Intermittently Connected Networks.”
Lecture Notes in Computer Science, 3126:239 – 254, August 2004.
108
[137] J. Burgess, B. Gallagher, D. Jensen, and B. N. Levine. “MaxProp: Routing for Vehicle-Based Disruption-
Tolerant Networking.” In Proc. of IEEE INFOCOM, 2006.
[138] Basu, S., Mukherjee, A., and Klivansky, S. “Time series models for internet traffic”. In INFOCOM (March
1996), vol. 2, pp. 611–620.
[139] Liu, C., Wu, K., and Tsao, M., “Energy efficient information collection with the ARIMA model in wireless
sensor networks”. In IEEE Global Telecommunications Conference GLOBECOM-05 (2005), vol. 5, pp. 2470–
2474.
[140] Brockwell, P. J., and Davis, R. Time Series: Theory and Methods. Springer-Verlag, New York, 1987.
[141] Herbert, D., Modelo-Howard, G., Perez-Toro, C., and Bagchi, S., Fault tolerant arima-based aggregation of
data in sensor networks. In IEEE International Conference on Dependable Systems and Networks (June 2007).
[142] Heinzelman, W., Chandrakasan, A., and Balakrishnan, H. An application-specific protocol architecture for
wireless microsensor networks. IEEE Transactions on Wireless Communications 1, 4 (October 2002), 660–670.
[143] Banerjee, A., Majumder, K., Dutta, P., and Mondal, K. Implementation of the behavior and structure of the
multihop mobile environment as a pushdown automata and birthand- death based statistical model. In
Mobile Computing and Networking (MOBICOMNET) (2004), pp. 45–51.
[144] Singh, J. P. and Dutta, P. 2009. Temporal behavior analysis of mobile ad hoc network with different
mobility patterns. In Proceedings of the international Conference on Advances in Computing, Communication
and Control (Mumbai, India, January 23 - 24, 2009). ICAC3 '09. ACM, New York, NY, 696-702.
[145] Box, G. E. P., and G. M. Jenkins, Time Series Analysis: Forecasting and Control, Holden-Day Inc., San
Francisco, Calif., 1970.
[146] Alparslan, N. D., and Khosrow, S. A generalized random mobility model for wireless ad hoc networks and
its analysis: One dimensional case. IEEE/ACM Transactions on Networking (TON) 15, 3 (June 2007), 602–
615.
[147] Hyytia, E., Lassila, P., and Virtamo, J. A markovian waypoint mobility model with application to hotspot
modeling. In IEEE International Conference on Communications (2006), vol. 3, pp. 979–986.
[148] Camp, T., Boleng, J., and Davies, V. A survey of mobility models for ad hoc network research. Wireless
Communications and Mobile Computing (WCMC): Special issue on Mobile Ad Hoc Networking: Research,
Trends and Applications 2, 5 (2002), 483–502.
[149] Minitab Inc, 2010. Minitab Statistical Software. Minitab Release 15. http://www.minitab.com/.
[150] Priestley, M. B. 1981 Spectral analysis and time series / M.B. Priestley Academic Press, London; New York.
[151] Sung Hwi Kim; Seung Min Oh; Ho Sung Park; Jeong Cheol Lee; Sang Ha Kim; , "Disjoint Multipath
Scheme with Hole Detouring Strategy in Wireless Sensor Networks," Vehicular Technology Conference (VTC
Fall), 2011 IEEE , vol., no., pp.1-5, 5-8 Sept. 2011
[152] A. Clementi, P. Penna and R. Silvestri, "Hardness Results for the Power Range Assignment Problem in
Packet Radio Networks," in proceedings of RANDOM/APPROX, 1999.
[153] L. Kirousis, E. Kranakis, D. Krizanc and A. Pelc, "Power Consumption in Packet Radio Networks,"
Theoret. Comput. Sci., pp. 289 - 305, 2000.
[154] S. Narayanaswamy, V. Kawadia, R. Sreenivas and P. Kumar, "Power control in ad hoc networks: Theory,
architecture, algorithm and implementation of the COMPOW protocol," in proceedings of European
Wireless, 2002.
[155] J. Pan, Y. T. Hou, L. Cai, Y. Shi and S. X. Shen, "Topology Control for Wireless Sensor Networks," in
proceedings of ACM Mobicom' 03, 2003.
[156] C. Schurgers, V. Tsiatsis, S. Ganeriwal and M. Srivastava, "Topology Management for Sensor Networks:
Exploiting Latency and Density," in proceedings of ACM Mobihoc, 2002.
[157] R. Zheng, J. C. Hou and L. Sha, "Asynchronous Wakeup for Ad Hoc Networks," in proceedings of ACM
Mobicom, 2003.
109
[158] N. Heo and P. K. Varshney, "An Intelligent Deployment and Clustering Algorithm for a Distributed
Mobile Sensor Network," in proceedings of IEEE International Conference on Systems, Man and
Cybernetics, 2003.
[159] A. Howard, M. J. Mataric and G. S. Sukhatme, "Mobile Sensor Network Deployment using Potential
Fields: A Distributed, Scalable Solution to the Area Coverage Problem," 2002.
[160] C. F. Huang and Y. C. Tseng, "The Coverage Problem in a Wireless Sensor Network," in proceedings of
ACM WSNA, 2003.
[161] G. Wang, G. Cao and T. L. Porta, "Movement-Assisted Sensor Deployment," in proceedings of IEEE
INFOCOM, 2004.
[162] X. Wang, G. Xing, Y. Zhang, C. Lu, R. Pless, et al., "Integrated Coverage and Connectivity Configuration
in Wireless Sensor Networks," in proceedings of ACM SenSys, 2003.
[163] T. Yan, T. He and J. Stankovic, "Differentiated Surveillance for Sensor Networks," in proceedings of ACM
SenSys, 2003.
[164] F. Ye, G. Zhong, S. Lu and L. Zhang, "PEAS: A Robust Energy Conserving Protocol for Long-lived Sensor
Networks," in proceedings of International Conference on Distributed Computing Systems (ICDCS), 2003.
[165] P. Santi and D. Blough, "The Critical Transmitting Range for Connectivity in Sparse Wireless Ad Hoc
Networks," IEEE Trans. on Mobile Computing, 2003.
[166] M. Li and B. Yang. "A survey on topology issues in wireless sensor network". In Proceedings of the 2006
International Conference on Wireless Networks, page 503, Las Vegas, Nevada, USA, 2006.
[167] Q. Cao, T. Abdelzaher, T. He and J. Stankovic, "Towards Optimal Sleep Scheduling in Sensor Networks
for Rare-Event Detection," in proceedings of IPSN, 2005.
[168] H. Zhang and J. Hou, "Maintaining Sensing Coverage and Connectivity in Large Sensor Networks,"
Department of Computer Science, UIUC UIUCDCS-R-2003-2351, 2003.
[169] M. A. Batalin and G. S. Sukhtame, "Coverage, Exploration and Deployment by a Mobile Robot and
Communication Network," Telecommunication Systems Journal, Special Issue on Wireless Sensor
Networks, vol. 26(2), pp. 181 - 196, 2004.
[170] G. Wang, G. Cao and T. L. Porta, "A Bidding Protocol for Deploying Mobile Sensors," in proceedings of
IEEE International Conference on Network Protocol (ICNP), 2003.
[171] M. Marina and S. Das, "Routing Performance in the Presence of Unidirectional Links in Multi-hop Wireless
Networks," in proceedings of ACM Mobihoc, 2002.
[172] Ferrero, R.; Gandino, F.; , "Degree Distribution of Unit Disk Graphs with Uniformly Deployed Nodes on
a Rectangular Surface," Broadband and Wireless Computing, Communication and Applications (BWCCA),
2011 International Conference on , vol., no., pp.255-262, 26-28 Oct. 2011
[173] Hekmat, R.; Van Mieghem, P.; , "Degree Distribution and Hop Count in Wireless Ad-hoc
Networks," Networks, 2003. ICON2003. The 11th IEEE International Conference on, vol., no., pp. 603- 609,
28 Sept.-1 Oct. 2003
[174] Lifang Guo; Harfoush, K.; Huimin Xu; , "Distribution of the Node Degree in MANETs," Next Generation
Mobile Applications, Services and Technologies (NGMAST), 2010 Fourth International Conference on , vol., no.,
pp.162-167, 27-29 July 2010.